diff --git a/vertical-pod-autoscaler/e2e/v1beta1/admission_controller.go b/vertical-pod-autoscaler/e2e/v1beta1/admission_controller.go
index a250f3fdd5bb..d40687bb7fca 100644
--- a/vertical-pod-autoscaler/e2e/v1beta1/admission_controller.go
+++ b/vertical-pod-autoscaler/e2e/v1beta1/admission_controller.go
@@ -63,7 +63,9 @@ var _ = AdmissionControllerE2eDescribe("Admission-controller", func() {
 		d := NewHamsterDeploymentWithGuaranteedResources(f, ParseQuantityOrDie("100m") /*cpu*/, ParseQuantityOrDie("100Mi") /*memory*/)
 
 		ginkgo.By("Setting up a VPA CRD")
-		vpaCRD := NewVPA(f, "hamster-vpa", hamsterTargetRef)
+		vpaCRD := NewVPA(f, "hamster-vpa", &metav1.LabelSelector{
+			MatchLabels: d.Spec.Template.Labels,
+		})
 		vpaCRD.Status.Recommendation = &vpa_types.RecommendedPodResources{
 			ContainerRecommendations: []vpa_types.RecommendedContainerResources{{
 				ContainerName: "hamster",
@@ -88,6 +90,123 @@ var _ = AdmissionControllerE2eDescribe("Admission-controller", func() {
 		}
 	})
 
+	ginkgo.It("keeps limits to request ratio constant", func() {
+		d := NewHamsterDeploymentWithResourcesAndLimits(f,
+			ParseQuantityOrDie("100m") /*cpu request*/, ParseQuantityOrDie("100Mi"), /*memory request*/
+			ParseQuantityOrDie("150m") /*cpu limit*/, ParseQuantityOrDie("200Mi") /*memory limit*/)
+
+		ginkgo.By("Setting up a VPA CRD")
+		vpaCRD := NewVPA(f, "hamster-vpa", &metav1.LabelSelector{
+			MatchLabels: d.Spec.Template.Labels,
+		})
+		vpaCRD.Status.Recommendation = &vpa_types.RecommendedPodResources{
+			ContainerRecommendations: []vpa_types.RecommendedContainerResources{{
+				ContainerName: "hamster",
+				Target: apiv1.ResourceList{
+					apiv1.ResourceCPU:    ParseQuantityOrDie("250m"),
+					apiv1.ResourceMemory: ParseQuantityOrDie("200Mi"),
+				},
+			}},
+		}
+		InstallVPA(f, vpaCRD)
+
+		ginkgo.By("Setting up a hamster deployment")
+		podList := startDeploymentPods(f, d)
+
+		// Originally Pods had 100m CPU, 100Mi of memory, but admission controller
+		// should change it to 250m CPU and 200Mi of memory. Limits to request ratio should stay unchanged.
+		for _, pod := range podList.Items {
+			gomega.Expect(*pod.Spec.Containers[0].Resources.Requests.Cpu()).To(gomega.Equal(ParseQuantityOrDie("250m")))
+			gomega.Expect(*pod.Spec.Containers[0].Resources.Requests.Memory()).To(gomega.Equal(ParseQuantityOrDie("200Mi")))
+			gomega.Expect(float64(pod.Spec.Containers[0].Resources.Limits.Cpu().MilliValue()) / float64(pod.Spec.Containers[0].Resources.Requests.Cpu().MilliValue())).To(gomega.BeNumerically("~", 1.5))
+			gomega.Expect(float64(pod.Spec.Containers[0].Resources.Limits.Memory().Value()) / float64(pod.Spec.Containers[0].Resources.Requests.Memory().Value())).To(gomega.BeNumerically("~", 2.))
+		}
+	})
+
+	ginkgo.It("caps request according to max limit set in LimitRange", func() {
+		d := NewHamsterDeploymentWithResourcesAndLimits(f,
+			ParseQuantityOrDie("100m") /*cpu request*/, ParseQuantityOrDie("100Mi"), /*memory request*/
+			ParseQuantityOrDie("150m") /*cpu limit*/, ParseQuantityOrDie("200Mi") /*memory limit*/)
+
+		ginkgo.By("Setting up a VPA CRD")
+		vpaCRD := NewVPA(f, "hamster-vpa", &metav1.LabelSelector{
+			MatchLabels: d.Spec.Template.Labels,
+		})
+		vpaCRD.Status.Recommendation = &vpa_types.RecommendedPodResources{
+			ContainerRecommendations: []vpa_types.RecommendedContainerResources{{
+				ContainerName: "hamster",
+				Target: apiv1.ResourceList{
+					apiv1.ResourceCPU:    ParseQuantityOrDie("250m"),
+					apiv1.ResourceMemory: ParseQuantityOrDie("200Mi"),
+				},
+			}},
+		}
+		InstallVPA(f, vpaCRD)
+
+		// Max CPU limit is 300m and ratio is 1.5, so max request is 200m, while
+		// recommendation is 250m
+		// Max memory limit is 1T and ratio is 2., so max request is 0.5T
+		InstallLimitRangeWithMax(f, "300m", "1T")
+
+		ginkgo.By("Setting up a hamster deployment")
+		podList := startDeploymentPods(f, d)
+
+		// Originally Pods had 100m CPU, 100Mi of memory, but admission controller
+		// should change it to 200m CPU (as this is the recommendation
+		// capped according to max limit in LimitRange) and 200Mi of memory,
+		// which is uncapped. Limit to request ratio should stay unchanged.
+		for _, pod := range podList.Items {
+			gomega.Expect(*pod.Spec.Containers[0].Resources.Requests.Cpu()).To(gomega.Equal(ParseQuantityOrDie("200m")))
+			gomega.Expect(*pod.Spec.Containers[0].Resources.Requests.Memory()).To(gomega.Equal(ParseQuantityOrDie("200Mi")))
+			gomega.Expect(*pod.Spec.Containers[0].Resources.Limits.Cpu()).To(gomega.BeNumerically("<=", ParseQuantityOrDie("300m")))
+			gomega.Expect(*pod.Spec.Containers[0].Resources.Limits.Memory()).To(gomega.BeNumerically("<=", ParseQuantityOrDie("1T")))
+			gomega.Expect(float64(pod.Spec.Containers[0].Resources.Limits.Cpu().MilliValue()) / float64(pod.Spec.Containers[0].Resources.Requests.Cpu().MilliValue())).To(gomega.BeNumerically("~", 1.5))
+			gomega.Expect(float64(pod.Spec.Containers[0].Resources.Limits.Memory().Value()) / float64(pod.Spec.Containers[0].Resources.Requests.Memory().Value())).To(gomega.BeNumerically("~", 2.))
+		}
+	})
+
+	ginkgo.It("raises request according to min limit set in LimitRange", func() {
+		d := NewHamsterDeploymentWithResourcesAndLimits(f,
+			ParseQuantityOrDie("100m") /*cpu request*/, ParseQuantityOrDie("200Mi"), /*memory request*/
+			ParseQuantityOrDie("150m") /*cpu limit*/, ParseQuantityOrDie("400Mi") /*memory limit*/)
+
+		ginkgo.By("Setting up a VPA CRD")
+		vpaCRD := NewVPA(f, "hamster-vpa", &metav1.LabelSelector{
+			MatchLabels: d.Spec.Template.Labels,
+		})
+		vpaCRD.Status.Recommendation = &vpa_types.RecommendedPodResources{
+			ContainerRecommendations: []vpa_types.RecommendedContainerResources{{
+				ContainerName: "hamster",
+				Target: apiv1.ResourceList{
+					apiv1.ResourceCPU:    ParseQuantityOrDie("250m"),
+					apiv1.ResourceMemory: ParseQuantityOrDie("100Mi"), // memory is downscaled
+				},
+			}},
+		}
+		InstallVPA(f, vpaCRD)
+
+		// Min CPU limit is 75m and ratio is 1.5, so min request is 50m
+		// Min memory limit is 250Mi and ratio is 2., so min request is 125Mi, while
+		// recommendation is 100Mi.
+		InstallLimitRangeWithMin(f, "75m", "250Mi")
+
+		ginkgo.By("Setting up a hamster deployment")
+		podList := startDeploymentPods(f, d)
+
+		// Originally Pods had 100m CPU, 200Mi of memory, but admission controller
+		// should change it to 250m CPU and 125Mi of memory, since this is the lowest
+		// request that limitrange allows.
+		// Limit to request ratio should stay unchanged.
+		for _, pod := range podList.Items {
+			gomega.Expect(*pod.Spec.Containers[0].Resources.Requests.Cpu()).To(gomega.Equal(ParseQuantityOrDie("250m")))
+			gomega.Expect(*pod.Spec.Containers[0].Resources.Requests.Memory()).To(gomega.Equal(ParseQuantityOrDie("125Mi")))
+			gomega.Expect(*pod.Spec.Containers[0].Resources.Limits.Cpu()).To(gomega.BeNumerically(">=", ParseQuantityOrDie("75m")))
+			gomega.Expect(*pod.Spec.Containers[0].Resources.Limits.Memory()).To(gomega.BeNumerically(">=", ParseQuantityOrDie("250Mi")))
+			gomega.Expect(float64(pod.Spec.Containers[0].Resources.Limits.Cpu().MilliValue()) / float64(pod.Spec.Containers[0].Resources.Requests.Cpu().MilliValue())).To(gomega.BeNumerically("~", 1.5))
+			gomega.Expect(float64(pod.Spec.Containers[0].Resources.Limits.Memory().Value()) / float64(pod.Spec.Containers[0].Resources.Requests.Memory().Value())).To(gomega.BeNumerically("~", 2.))
+		}
+	})
+
 	ginkgo.It("caps request to max set in VPA", func() {
 		d := NewHamsterDeploymentWithResources(f, ParseQuantityOrDie("100m") /*cpu*/, ParseQuantityOrDie("100Mi") /*memory*/)
 
diff --git a/vertical-pod-autoscaler/e2e/v1beta1/common.go b/vertical-pod-autoscaler/e2e/v1beta1/common.go
index 98db1b4d0e08..4780be450db4 100644
--- a/vertical-pod-autoscaler/e2e/v1beta1/common.go
+++ b/vertical-pod-autoscaler/e2e/v1beta1/common.go
@@ -136,6 +136,17 @@ func NewHamsterDeploymentWithGuaranteedResources(f *framework.Framework, cpuQuan
 	return d
 }
 
+// NewHamsterDeploymentWithResourcesAndLimits creates a simple hamster deployment with specific
+// resource requests and limits for e2e test purposes.
+func NewHamsterDeploymentWithResourcesAndLimits(f *framework.Framework, cpuQuantityRequest, memoryQuantityRequest, cpuQuantityLimit, memoryQuantityLimit resource.Quantity) *appsv1.Deployment {
+	d := NewHamsterDeploymentWithResources(f, cpuQuantityRequest, memoryQuantityRequest)
+	d.Spec.Template.Spec.Containers[0].Resources.Limits = apiv1.ResourceList{
+		apiv1.ResourceCPU:    cpuQuantityLimit,
+		apiv1.ResourceMemory: memoryQuantityLimit,
+	}
+	return d
+}
+
 // GetHamsterPods returns running hamster pods (matched by hamsterLabels)
 func GetHamsterPods(f *framework.Framework) (*apiv1.PodList, error) {
 	label := labels.SelectorFromSet(labels.Set(hamsterLabels))
@@ -342,3 +353,61 @@ func WaitForConditionPresent(c *vpa_clientset.Clientset, vpa *vpa_types.Vertical
 		return false
 	})
 }
+
+func installLimitRange(f *framework.Framework, minCpuLimit, minMemoryLimit, maxCpuLimit, maxMemoryLimit *resource.Quantity) {
+	lr := &apiv1.LimitRange{
+		ObjectMeta: metav1.ObjectMeta{
+			Namespace: f.Namespace.Name,
+			Name:      "hamster-lr",
+		},
+		Spec: apiv1.LimitRangeSpec{
+			Limits: []apiv1.LimitRangeItem{},
+		},
+	}
+
+	if maxMemoryLimit != nil || maxCpuLimit != nil {
+		lrItem := apiv1.LimitRangeItem{
+			Type: apiv1.LimitTypeContainer,
+			Max:  apiv1.ResourceList{},
+		}
+		if maxCpuLimit != nil {
+			lrItem.Max[apiv1.ResourceCPU] = *maxCpuLimit
+		}
+		if maxMemoryLimit != nil {
+			lrItem.Max[apiv1.ResourceMemory] = *maxMemoryLimit
+		}
+		lr.Spec.Limits = append(lr.Spec.Limits, lrItem)
+	}
+
+	if minMemoryLimit != nil || minCpuLimit != nil {
+		lrItem := apiv1.LimitRangeItem{
+			Type: apiv1.LimitTypeContainer,
+			Max:  apiv1.ResourceList{},
+		}
+		if minCpuLimit != nil {
+			lrItem.Min[apiv1.ResourceCPU] = *minCpuLimit
+		}
+		if minMemoryLimit != nil {
+			lrItem.Min[apiv1.ResourceMemory] = *minMemoryLimit
+		}
+		lr.Spec.Limits = append(lr.Spec.Limits, lrItem)
+	}
+	_, err := f.ClientSet.CoreV1().LimitRanges(f.Namespace.Name).Create(lr)
+	gomega.Expect(err).NotTo(gomega.HaveOccurred())
+}
+
+// InstallLimitRangeWithMax installs a LimitRange with a maximum limit for CPU and memory.
+func InstallLimitRangeWithMax(f *framework.Framework, maxCpuLimit, maxMemoryLimit string) {
+	ginkgo.By(fmt.Sprintf("Setting up LimitRange with max limits - CPU: %v, memory: %v", maxCpuLimit, maxMemoryLimit))
+	maxCpuLimitQuantity := ParseQuantityOrDie(maxCpuLimit)
+	maxMemoryLimitQuantity := ParseQuantityOrDie(maxMemoryLimit)
+	installLimitRange(f, nil, nil, &maxCpuLimitQuantity, &maxMemoryLimitQuantity)
+}
+
+// InstallLimitRangeWithMin installs a LimitRange with a minimum limit for CPU and memory.
+func InstallLimitRangeWithMin(f *framework.Framework, minCpuLimit, minMemoryLimit string) {
+	ginkgo.By(fmt.Sprintf("Setting up LimitRange with min limits - CPU: %v, memory: %v", minCpuLimit, minMemoryLimit))
+	minCpuLimitQuantity := ParseQuantityOrDie(minCpuLimit)
+	minMemoryLimitQuantity := ParseQuantityOrDie(minMemoryLimit)
+	installLimitRange(f, &minCpuLimitQuantity, &minMemoryLimitQuantity, nil, nil)
+}
diff --git a/vertical-pod-autoscaler/e2e/v1beta1/updater.go b/vertical-pod-autoscaler/e2e/v1beta1/updater.go
index 33b66e1d49d2..d61949fd5ee3 100644
--- a/vertical-pod-autoscaler/e2e/v1beta1/updater.go
+++ b/vertical-pod-autoscaler/e2e/v1beta1/updater.go
@@ -98,6 +98,44 @@ var _ = UpdaterE2eDescribe("Updater", func() {
 		gomega.Expect(err).NotTo(gomega.HaveOccurred())
 		gomega.Expect(evictedCount >= permissiveMaxUnavailable).To(gomega.BeTrue())
 	})
+
+	ginkgo.It("observes max in LimitRange", func() {
+		ginkgo.By("Setting up a hamster deployment")
+		d := NewHamsterDeploymentWithResourcesAndLimits(f,
+			ParseQuantityOrDie("100m") /*cpu request*/, ParseQuantityOrDie("200Mi"), /*memory request*/
+			ParseQuantityOrDie("300m") /*cpu limit*/, ParseQuantityOrDie("400Mi") /*memory limit*/)
+		podList := startDeploymentPods(f, d)
+
+		ginkgo.By("Setting up a VPA CRD")
+		SetupVPA(f, "200m", vpa_types.UpdateModeAuto)
+
+		// Max CPU limit is 300m and ratio is 3., so max request is 100m, while
+		// recommendation is 200m
+		// Max memory limit is 1T and ratio is 2., so max request is 0.5T
+		InstallLimitRangeWithMax(f, "300m", "1T")
+
+		ginkgo.By(fmt.Sprintf("Waiting for pods to be evicted, hoping it won't happen, sleep for %s", VpaEvictionTimeout.String()))
+		CheckNoPodsEvicted(f, MakePodSet(podList))
+	})
+
+	ginkgo.It("observes min in LimitRange", func() {
+		ginkgo.By("Setting up a hamster deployment")
+		d := NewHamsterDeploymentWithResourcesAndLimits(f,
+			ParseQuantityOrDie("100m") /*cpu request*/, ParseQuantityOrDie("200Mi"), /*memory request*/
+			ParseQuantityOrDie("300m") /*cpu limit*/, ParseQuantityOrDie("400Mi") /*memory limit*/)
+		podList := startDeploymentPods(f, d)
+
+		ginkgo.By("Setting up a VPA CRD")
+		SetupVPA(f, "50m", vpa_types.UpdateModeAuto)
+
+		// Min CPU limit is 300m and ratio is 3., so min request is 100m, while
+		// recommendation is 200m
+		// Min memory limit is 0 and ratio is 2., so min request is 0
+		InstallLimitRangeWithMin(f, "300m", "0")
+
+		ginkgo.By(fmt.Sprintf("Waiting for pods to be evicted, hoping it won't happen, sleep for %s", VpaEvictionTimeout.String()))
+		CheckNoPodsEvicted(f, MakePodSet(podList))
+	})
 })
 
 func testEvictsPods(f *framework.Framework, controllerKind string) {
diff --git a/vertical-pod-autoscaler/e2e/v1beta2/admission_controller.go b/vertical-pod-autoscaler/e2e/v1beta2/admission_controller.go
index cb7f4d92074f..98d2306c4a78 100644
--- a/vertical-pod-autoscaler/e2e/v1beta2/admission_controller.go
+++ b/vertical-pod-autoscaler/e2e/v1beta2/admission_controller.go
@@ -85,6 +85,117 @@ var _ = AdmissionControllerE2eDescribe("Admission-controller", func() {
 		}
 	})
 
+	ginkgo.It("keeps limits to request ratio constant", func() {
+		d := NewHamsterDeploymentWithResourcesAndLimits(f,
+			ParseQuantityOrDie("100m") /*cpu request*/, ParseQuantityOrDie("100Mi"), /*memory request*/
+			ParseQuantityOrDie("150m") /*cpu limit*/, ParseQuantityOrDie("200Mi") /*memory limit*/)
+
+		ginkgo.By("Setting up a VPA CRD")
+		vpaCRD := NewVPA(f, "hamster-vpa", hamsterTargetRef)
+		vpaCRD.Status.Recommendation = &vpa_types.RecommendedPodResources{
+			ContainerRecommendations: []vpa_types.RecommendedContainerResources{{
+				ContainerName: "hamster",
+				Target: apiv1.ResourceList{
+					apiv1.ResourceCPU:    ParseQuantityOrDie("250m"),
+					apiv1.ResourceMemory: ParseQuantityOrDie("200Mi"),
+				},
+			}},
+		}
+		InstallVPA(f, vpaCRD)
+
+		ginkgo.By("Setting up a hamster deployment")
+		podList := startDeploymentPods(f, d)
+
+		// Originally Pods had 100m CPU, 100Mi of memory, but admission controller
+		// should change it to 250m CPU and 200Mi of memory. Limits to request ratio should stay unchanged.
+		for _, pod := range podList.Items {
+			gomega.Expect(*pod.Spec.Containers[0].Resources.Requests.Cpu()).To(gomega.Equal(ParseQuantityOrDie("250m")))
+			gomega.Expect(*pod.Spec.Containers[0].Resources.Requests.Memory()).To(gomega.Equal(ParseQuantityOrDie("200Mi")))
+			gomega.Expect(float64(pod.Spec.Containers[0].Resources.Limits.Cpu().MilliValue()) / float64(pod.Spec.Containers[0].Resources.Requests.Cpu().MilliValue())).To(gomega.BeNumerically("~", 1.5))
+			gomega.Expect(float64(pod.Spec.Containers[0].Resources.Limits.Memory().Value()) / float64(pod.Spec.Containers[0].Resources.Requests.Memory().Value())).To(gomega.BeNumerically("~", 2.))
+		}
+	})
+
+	ginkgo.It("caps request according to max limit set in LimitRange", func() {
+		d := NewHamsterDeploymentWithResourcesAndLimits(f,
+			ParseQuantityOrDie("100m") /*cpu request*/, ParseQuantityOrDie("100Mi"), /*memory request*/
+			ParseQuantityOrDie("150m") /*cpu limit*/, ParseQuantityOrDie("200Mi") /*memory limit*/)
+
+		ginkgo.By("Setting up a VPA CRD")
+		vpaCRD := NewVPA(f, "hamster-vpa", hamsterTargetRef)
+		vpaCRD.Status.Recommendation = &vpa_types.RecommendedPodResources{
+			ContainerRecommendations: []vpa_types.RecommendedContainerResources{{
+				ContainerName: "hamster",
+				Target: apiv1.ResourceList{
+					apiv1.ResourceCPU:    ParseQuantityOrDie("250m"),
+					apiv1.ResourceMemory: ParseQuantityOrDie("200Mi"),
+				},
+			}},
+		}
+		InstallVPA(f, vpaCRD)
+
+		// Max CPU limit is 300m and ratio is 1.5, so max request is 200m, while
+		// recommendation is 250m
+		// Max memory limit is 1T and ratio is 2., so max request is 0.5T
+		InstallLimitRangeWithMax(f, "300m", "1T")
+
+		ginkgo.By("Setting up a hamster deployment")
+		podList := startDeploymentPods(f, d)
+
+		// Originally Pods had 100m CPU, 100Mi of memory, but admission controller
+		// should change it to 200m CPU (as this is the recommendation
+		// capped according to max limit in LimitRange) and 200Mi of memory,
+		// which is uncapped. Limit to request ratio should stay unchanged.
+		for _, pod := range podList.Items {
+			gomega.Expect(*pod.Spec.Containers[0].Resources.Requests.Cpu()).To(gomega.Equal(ParseQuantityOrDie("200m")))
+			gomega.Expect(*pod.Spec.Containers[0].Resources.Requests.Memory()).To(gomega.Equal(ParseQuantityOrDie("200Mi")))
+			gomega.Expect(*pod.Spec.Containers[0].Resources.Limits.Cpu()).To(gomega.BeNumerically("<=", ParseQuantityOrDie("300m")))
+			gomega.Expect(*pod.Spec.Containers[0].Resources.Limits.Memory()).To(gomega.BeNumerically("<=", ParseQuantityOrDie("1T")))
+			gomega.Expect(float64(pod.Spec.Containers[0].Resources.Limits.Cpu().MilliValue()) / float64(pod.Spec.Containers[0].Resources.Requests.Cpu().MilliValue())).To(gomega.BeNumerically("~", 1.5))
+			gomega.Expect(float64(pod.Spec.Containers[0].Resources.Limits.Memory().Value()) / float64(pod.Spec.Containers[0].Resources.Requests.Memory().Value())).To(gomega.BeNumerically("~", 2.))
+		}
+	})
+
+	ginkgo.It("raises request according to min limit set in LimitRange", func() {
+		d := NewHamsterDeploymentWithResourcesAndLimits(f,
+			ParseQuantityOrDie("100m") /*cpu request*/, ParseQuantityOrDie("200Mi"), /*memory request*/
+			ParseQuantityOrDie("150m") /*cpu limit*/, ParseQuantityOrDie("400Mi") /*memory limit*/)
+
+		ginkgo.By("Setting up a VPA CRD")
+		vpaCRD := NewVPA(f, "hamster-vpa", hamsterTargetRef)
+		vpaCRD.Status.Recommendation = &vpa_types.RecommendedPodResources{
+			ContainerRecommendations: []vpa_types.RecommendedContainerResources{{
+				ContainerName: "hamster",
+				Target: apiv1.ResourceList{
+					apiv1.ResourceCPU:    ParseQuantityOrDie("250m"),
+					apiv1.ResourceMemory: ParseQuantityOrDie("100Mi"), // memory is downscaled
+				},
+			}},
+		}
+		InstallVPA(f, vpaCRD)
+
+		// Min CPU limit is 75m and ratio is 1.5, so min request is 50m
+		// Min memory limit is 250Mi and ratio is 2., so min request is 125Mi, while
+		// recommendation is 100Mi.
+		InstallLimitRangeWithMin(f, "75m", "250Mi")
+
+		ginkgo.By("Setting up a hamster deployment")
+		podList := startDeploymentPods(f, d)
+
+		// Originally Pods had 100m CPU, 200Mi of memory, but admission controller
+		// should change it to 250m CPU and 125Mi of memory, since this is the lowest
+		// request that limitrange allows.
+		// Limit to request ratio should stay unchanged.
+		for _, pod := range podList.Items {
+			gomega.Expect(*pod.Spec.Containers[0].Resources.Requests.Cpu()).To(gomega.Equal(ParseQuantityOrDie("250m")))
+			gomega.Expect(*pod.Spec.Containers[0].Resources.Requests.Memory()).To(gomega.Equal(ParseQuantityOrDie("125Mi")))
+			gomega.Expect(*pod.Spec.Containers[0].Resources.Limits.Cpu()).To(gomega.BeNumerically(">=", ParseQuantityOrDie("75m")))
+			gomega.Expect(*pod.Spec.Containers[0].Resources.Limits.Memory()).To(gomega.BeNumerically(">=", ParseQuantityOrDie("250Mi")))
+			gomega.Expect(float64(pod.Spec.Containers[0].Resources.Limits.Cpu().MilliValue()) / float64(pod.Spec.Containers[0].Resources.Requests.Cpu().MilliValue())).To(gomega.BeNumerically("~", 1.5))
+			gomega.Expect(float64(pod.Spec.Containers[0].Resources.Limits.Memory().Value()) / float64(pod.Spec.Containers[0].Resources.Requests.Memory().Value())).To(gomega.BeNumerically("~", 2.))
+		}
+	})
+
 	ginkgo.It("caps request to max set in VPA", func() {
 		d := NewHamsterDeploymentWithResources(f, ParseQuantityOrDie("100m") /*cpu*/, ParseQuantityOrDie("100Mi") /*memory*/)
 
diff --git a/vertical-pod-autoscaler/e2e/v1beta2/common.go b/vertical-pod-autoscaler/e2e/v1beta2/common.go
index 93d33906e311..b3980f7c9b83 100644
--- a/vertical-pod-autoscaler/e2e/v1beta2/common.go
+++ b/vertical-pod-autoscaler/e2e/v1beta2/common.go
@@ -143,6 +143,17 @@ func NewHamsterDeploymentWithGuaranteedResources(f *framework.Framework, cpuQuan
 	return d
 }
 
+// NewHamsterDeploymentWithResourcesAndLimits creates a simple hamster deployment with specific
+// resource requests and limits for e2e test purposes.
+func NewHamsterDeploymentWithResourcesAndLimits(f *framework.Framework, cpuQuantityRequest, memoryQuantityRequest, cpuQuantityLimit, memoryQuantityLimit resource.Quantity) *appsv1.Deployment {
+	d := NewHamsterDeploymentWithResources(f, cpuQuantityRequest, memoryQuantityRequest)
+	d.Spec.Template.Spec.Containers[0].Resources.Limits = apiv1.ResourceList{
+		apiv1.ResourceCPU:    cpuQuantityLimit,
+		apiv1.ResourceMemory: memoryQuantityLimit,
+	}
+	return d
+}
+
 // GetHamsterPods returns running hamster pods (matched by hamsterLabels)
 func GetHamsterPods(f *framework.Framework) (*apiv1.PodList, error) {
 	label := labels.SelectorFromSet(labels.Set(hamsterLabels))
@@ -335,3 +346,61 @@ func WaitForRecommendationPresent(c *vpa_clientset.Clientset, vpa *vpa_types.Ver
 		return vpa.Status.Recommendation != nil && len(vpa.Status.Recommendation.ContainerRecommendations) != 0
 	})
 }
+
+func installLimitRange(f *framework.Framework, minCpuLimit, minMemoryLimit, maxCpuLimit, maxMemoryLimit *resource.Quantity) {
+	lr := &apiv1.LimitRange{
+		ObjectMeta: metav1.ObjectMeta{
+			Namespace: f.Namespace.Name,
+			Name:      "hamster-lr",
+		},
+		Spec: apiv1.LimitRangeSpec{
+			Limits: []apiv1.LimitRangeItem{},
+		},
+	}
+
+	if maxMemoryLimit != nil || maxCpuLimit != nil {
+		lrItem := apiv1.LimitRangeItem{
+			Type: apiv1.LimitTypeContainer,
+			Max:  apiv1.ResourceList{},
+		}
+		if maxCpuLimit != nil {
+			lrItem.Max[apiv1.ResourceCPU] = *maxCpuLimit
+		}
+		if maxMemoryLimit != nil {
+			lrItem.Max[apiv1.ResourceMemory] = *maxMemoryLimit
+		}
+		lr.Spec.Limits = append(lr.Spec.Limits, lrItem)
+	}
+
+	if minMemoryLimit != nil || minCpuLimit != nil {
+		lrItem := apiv1.LimitRangeItem{
+			Type: apiv1.LimitTypeContainer,
+			Max:  apiv1.ResourceList{},
+		}
+		if minCpuLimit != nil {
+			lrItem.Min[apiv1.ResourceCPU] = *minCpuLimit
+		}
+		if minMemoryLimit != nil {
+			lrItem.Min[apiv1.ResourceMemory] = *minMemoryLimit
+		}
+		lr.Spec.Limits = append(lr.Spec.Limits, lrItem)
+	}
+	_, err := f.ClientSet.CoreV1().LimitRanges(f.Namespace.Name).Create(lr)
+	gomega.Expect(err).NotTo(gomega.HaveOccurred())
+}
+
+// InstallLimitRangeWithMax installs a LimitRange with a maximum limit for CPU and memory.
+func InstallLimitRangeWithMax(f *framework.Framework, maxCpuLimit, maxMemoryLimit string) {
+	ginkgo.By(fmt.Sprintf("Setting up LimitRange with max limits - CPU: %v, memory: %v", maxCpuLimit, maxMemoryLimit))
+	maxCpuLimitQuantity := ParseQuantityOrDie(maxCpuLimit)
+	maxMemoryLimitQuantity := ParseQuantityOrDie(maxMemoryLimit)
+	installLimitRange(f, nil, nil, &maxCpuLimitQuantity, &maxMemoryLimitQuantity)
+}
+
+// InstallLimitRangeWithMin installs a LimitRange with a minimum limit for CPU and memory.
+func InstallLimitRangeWithMin(f *framework.Framework, minCpuLimit, minMemoryLimit string) {
+	ginkgo.By(fmt.Sprintf("Setting up LimitRange with min limits - CPU: %v, memory: %v", minCpuLimit, minMemoryLimit))
+	minCpuLimitQuantity := ParseQuantityOrDie(minCpuLimit)
+	minMemoryLimitQuantity := ParseQuantityOrDie(minMemoryLimit)
+	installLimitRange(f, &minCpuLimitQuantity, &minMemoryLimitQuantity, nil, nil)
+}
diff --git a/vertical-pod-autoscaler/e2e/v1beta2/updater.go b/vertical-pod-autoscaler/e2e/v1beta2/updater.go
index 67f9edb202d6..89b8492a726b 100644
--- a/vertical-pod-autoscaler/e2e/v1beta2/updater.go
+++ b/vertical-pod-autoscaler/e2e/v1beta2/updater.go
@@ -119,6 +119,44 @@ var _ = UpdaterE2eDescribe("Updater", func() {
 		gomega.Expect(err).NotTo(gomega.HaveOccurred())
 		gomega.Expect(evictedCount >= permissiveMaxUnavailable).To(gomega.BeTrue())
 	})
+
+	ginkgo.It("observes max in LimitRange", func() {
+		ginkgo.By("Setting up a hamster deployment")
+		d := NewHamsterDeploymentWithResourcesAndLimits(f,
+			ParseQuantityOrDie("100m") /*cpu request*/, ParseQuantityOrDie("200Mi"), /*memory request*/
+			ParseQuantityOrDie("300m") /*cpu limit*/, ParseQuantityOrDie("400Mi") /*memory limit*/)
+		podList := startDeploymentPods(f, d)
+
+		ginkgo.By("Setting up a VPA CRD")
+		SetupVPA(f, "200m", vpa_types.UpdateModeAuto, hamsterTargetRef)
+
+		// Max CPU limit is 300m and ratio is 3., so max request is 100m, while
+		// recommendation is 200m
+		// Max memory limit is 1T and ratio is 2., so max request is 0.5T
+		InstallLimitRangeWithMax(f, "300m", "1T")
+
+		ginkgo.By(fmt.Sprintf("Waiting for pods to be evicted, hoping it won't happen, sleep for %s", VpaEvictionTimeout.String()))
+		CheckNoPodsEvicted(f, MakePodSet(podList))
+	})
+
+	ginkgo.It("observes min in LimitRange", func() {
+		ginkgo.By("Setting up a hamster deployment")
+		d := NewHamsterDeploymentWithResourcesAndLimits(f,
+			ParseQuantityOrDie("100m") /*cpu request*/, ParseQuantityOrDie("200Mi"), /*memory request*/
+			ParseQuantityOrDie("300m") /*cpu limit*/, ParseQuantityOrDie("400Mi") /*memory limit*/)
+		podList := startDeploymentPods(f, d)
+
+		ginkgo.By("Setting up a VPA CRD")
+		SetupVPA(f, "50m", vpa_types.UpdateModeAuto, hamsterTargetRef)
+
+		// Min CPU limit is 300m and ratio is 3., so min request is 100m, while
+		// recommendation is 200m
+		// Min memory limit is 0 and ratio is 2., so min request is 0
+		InstallLimitRangeWithMin(f, "300m", "0")
+
+		ginkgo.By(fmt.Sprintf("Waiting for pods to be evicted, hoping it won't happen, sleep for %s", VpaEvictionTimeout.String()))
+		CheckNoPodsEvicted(f, MakePodSet(podList))
+	})
 })
 
 func testEvictsPods(f *framework.Framework, controller *autoscaling.CrossVersionObjectReference) {