diff --git a/test/data/test_tensor_frame.py b/test/data/test_tensor_frame.py
index 6c283ecc..a0645b18 100644
--- a/test/data/test_tensor_frame.py
+++ b/test/data/test_tensor_frame.py
@@ -245,3 +245,11 @@ def test_custom_tf_get_col_feat():
     assert torch.equal(feat, feat_dict['numerical'][:, 0:1])
     feat = tf.get_col_feat('num_2')
     assert torch.equal(feat, feat_dict['numerical'][:, 1:2])
+
+
+def test_non_list_col_names_dict():
+    feat_dict = {torch_frame.categorical: torch.randint(0, 3, size=(10, 1))}
+    # Oops, user provided a single column name without wrapping it in a list:
+    col_names_dict = {torch_frame.categorical: 'cat_1'}
+    with pytest.raises(ValueError, match='must be a list of column names'):
+        TensorFrame(feat_dict, col_names_dict)
diff --git a/torch_frame/data/tensor_frame.py b/torch_frame/data/tensor_frame.py
index c9b9c56d..e88d8137 100644
--- a/torch_frame/data/tensor_frame.py
+++ b/torch_frame/data/tensor_frame.py
@@ -95,7 +95,13 @@ def validate(self) -> None:
         num_rows = self.num_rows
         empty_stypes: list[torch_frame.stype] = []
         for stype_name, feats in self.feat_dict.items():
-            num_cols = len(self.col_names_dict[stype_name])
+            col_names = self.col_names_dict[stype_name]
+            if not isinstance(col_names, list):
+                raise ValueError(
+                    f"col_names_dict[{stype_name}] must be a list of column "
+                    f"names.")
+
+            num_cols = len(col_names)
             if num_cols == 0:
                 empty_stypes.append(stype_name)