diff --git a/ldb/dataset.py b/ldb/dataset.py
index 5b966712..712ba7e9 100644
--- a/ldb/dataset.py
+++ b/ldb/dataset.py
@@ -314,18 +314,25 @@ def check_datasets_for_data_objects(
     error: bool = True,
 ) -> Iterator[str]:
     # collect data objects from datasets
-    ds_version_idents = get_all_dataset_version_identifiers(ldb_dir)
-    objects_in_datasets = {
-        d
-        for version_id in ds_version_idents.keys()
-        for d in get_collection(ldb_dir, version_id)
-    }
+    collection_identifiers = get_all_dataset_version_identifiers(ldb_dir)
+    obj_to_identifiers = defaultdict(list)
+    for collection_id, ds_identifiers in collection_identifiers.items():
+        if ds_identifiers:
+            for data_obj_hash in load_data_file(
+                get_hash_path(
+                    ldb_dir / InstanceDir.COLLECTIONS,
+                    collection_id,
+                ),
+            ):
+                obj_to_identifiers[data_obj_hash].extend(ds_identifiers)
     for data_obj_hash in data_object_hashes:
-        if data_obj_hash in objects_in_datasets:
+        obj_ds_identifiers = obj_to_identifiers.get(data_obj_hash)
+        if obj_ds_identifiers is not None:
             if error:
+                ds_ident_str = "\n".join(f"  {i}" for i in obj_ds_identifiers)
                 raise LDBException(
-                    f"Data object id:{data_obj_hash} is contained in a saved "
-                    "dataset",
+                    f"Data object id:{data_obj_hash} is contained in saved "
+                    f"datasets:\n{ds_ident_str}",
                 )
         else:
             yield data_obj_hash
@@ -342,15 +349,21 @@ def get_all_dataset_version_identifiers(ldb_dir: Path) -> Dict[str, List[str]]:
             "ab3...": [],
         }
     """
+    ds_version_dir = os.path.join(ldb_dir, InstanceDir.DATASET_VERSIONS)
+    collection_dir = os.path.join(ldb_dir, InstanceDir.COLLECTIONS)
     result = defaultdict(list)
     for dataset in iter_datasets(ldb_dir):
         for i, version_id in dataset.numbered_versions().items():
-            result[version_id].append(
+            version_path = get_hash_path(
+                Path(ds_version_dir),
+                version_id,
+            )
+            version_obj = DatasetVersion.parse(load_data_file(version_path))
+            result[version_obj.collection].append(
                 format_dataset_identifier(dataset.name, i),
             )
-    ds_version_dir = str(ldb_dir / InstanceDir.DATASET_VERSIONS)
-    for parent in os.listdir(ds_version_dir):
-        for filename in os.listdir(os.path.join(ds_version_dir, parent)):
+    for parent in os.listdir(collection_dir):
+        for filename in os.listdir(os.path.join(collection_dir, parent)):
             result[  # pylint: disable=pointless-statement
                 f"{parent}{filename}"
             ]
diff --git a/ldb/ds.py b/ldb/ds.py
index 7b2e97d7..65d67c3e 100644
--- a/ldb/ds.py
+++ b/ldb/ds.py
@@ -3,10 +3,20 @@
 from pathlib import Path
 from typing import Iterable, Iterator
 
-from ldb.dataset import iter_datasets
+from ldb.dataset import (
+    Dataset,
+    DatasetVersion,
+    get_all_dataset_version_identifiers,
+    iter_datasets,
+)
 from ldb.exceptions import DatasetNotFoundError
 from ldb.path import InstanceDir
-from ldb.utils import format_dataset_identifier, parse_dataset_identifier
+from ldb.utils import (
+    format_dataset_identifier,
+    get_hash_path,
+    load_data_file,
+    parse_dataset_identifier,
+)
 
 
 @dataclass
@@ -46,6 +56,8 @@ def print_ds_listings(
 
 def delete_datasets(ldb_dir: Path, ds_identifiers: Iterable[str]) -> None:
     ds_dir = os.path.join(ldb_dir, InstanceDir.DATASETS)
+    ds_version_dir = os.path.join(ldb_dir, InstanceDir.DATASET_VERSIONS)
+    collection_dir = os.path.join(ldb_dir, InstanceDir.COLLECTIONS)
     ds_info = []
     for ds_ident in ds_identifiers:
         name, version = parse_dataset_identifier(ds_ident)
@@ -63,11 +75,36 @@ def delete_datasets(ldb_dir: Path, ds_identifiers: Iterable[str]) -> None:
                 f"Dataset not found: {ds_ident}",
             )
 
+    collection_identifiers = get_all_dataset_version_identifiers(ldb_dir)
     for ds_ident, path in ds_info:
         try:
-            os.unlink(path)
+            dataset = Dataset.parse(load_data_file(Path(path)))
         except FileNotFoundError as exc:
             raise DatasetNotFoundError(
                 f"Dataset not found: {ds_ident}",
             ) from exc
+        for i, ds_version in dataset.numbered_versions().items():
+            version_path = get_hash_path(
+                Path(ds_version_dir),
+                ds_version,
+            )
+            version_obj = DatasetVersion.parse(load_data_file(version_path))
+            ds_version_ident = format_dataset_identifier(dataset.name, i)
+            refs = collection_identifiers[version_obj.collection]
+            refs.remove(ds_version_ident)
+            if not refs:
+                collection_path = str(
+                    get_hash_path(
+                        Path(collection_dir),
+                        version_obj.collection,
+                    ),
+                )
+                os.unlink(collection_path)
+                try:
+                    os.rmdir(
+                        os.path.split(collection_path.rstrip(os.path.sep))[0],
+                    )
+                except OSError:
+                    pass
+        os.unlink(path)
         print(f"Deleted {ds_ident}")