diff --git a/nvtabular/ops/categorify.py b/nvtabular/ops/categorify.py
index 556e2a005a..ebdbc5e45f 100644
--- a/nvtabular/ops/categorify.py
+++ b/nvtabular/ops/categorify.py
@@ -1704,7 +1704,10 @@ def _encode(
             codes = type(df)({"order": dispatch.arange(len(df), like_df=df)}, index=df.index)
 
         for cl, cr in zip(selection_l.names, selection_r.names):
-            if isinstance(df[cl].dropna().iloc[0], (np.ndarray, list)):
+            column_without_nans = df[cl].dropna()
+            if len(column_without_nans) and isinstance(
+                column_without_nans.iloc[0], (np.ndarray, list)
+            ):
                 ser = df[cl].copy()
                 codes[cl] = dispatch.flatten_list_column_values(ser).astype(value[cr].dtype)
             else:
diff --git a/tests/unit/ops/test_categorify.py b/tests/unit/ops/test_categorify.py
index 41a69ef346..5092bb7fa7 100644
--- a/tests/unit/ops/test_categorify.py
+++ b/tests/unit/ops/test_categorify.py
@@ -734,3 +734,18 @@ def test_categorify_inference():
     output_tensors = inference_op.transform(cats.input_columns, input_tensors)
     for key in input_tensors:
         assert output_tensors[key].dtype == np.dtype("int64")
+
+
+def test_categorify_transform_only_nans_column():
+    train_df = make_df({"cat_column": ["a", "a", "b", "c", np.nan]})
+    cat_features = ["cat_column"] >> nvt.ops.Categorify()
+    train_dataset = nvt.Dataset(train_df)
+
+    workflow = nvt.Workflow(cat_features)
+    workflow.fit(train_dataset)
+
+    inference_df = make_df({"cat_column": [np.nan] * 10})
+    inference_dataset = nvt.Dataset(inference_df)
+
+    output = workflow.transform(inference_dataset).compute()
+    assert len(output) == len(inference_df)