diff --git a/python/cuml/ensemble/randomforestclassifier.pyx b/python/cuml/ensemble/randomforestclassifier.pyx
index 78e8fd63b8..1772b9678e 100644
--- a/python/cuml/ensemble/randomforestclassifier.pyx
+++ b/python/cuml/ensemble/randomforestclassifier.pyx
@@ -145,6 +145,19 @@ class RandomForestClassifier(BaseRandomForestModel,
         reduce memory consumption.
       * While training the model for multi class classification problems,
         using deep trees or `max_features=1.0` provides better performance.
+      * Prediction of classes is currently different from how scikit-learn
+        predicts:
+          * scikit-learn predicts random forest classifiers by obtaining class
+            probabilities from each component tree, then averaging these class
+            probabilities over all the ensemble members, and finally resolving
+            to the label with highest probability as prediction.
+          * cuml random forest classifier prediction differs in that, each
+            component tree generates labels instead of class probabilities;
+            with the most frequent label over all the trees (the statistical
+            mode) resolved as prediction.
+        The above differences might cause marginal variations in accuracy in
+        tradeoff to better performance.
+        See: https://github.com/rapidsai/cuml/issues/3764
 
     Examples
     --------