diff --git a/tests/test_processing.py b/tests/test_processing.py
index fbe94ab..77b2aaa 100644
--- a/tests/test_processing.py
+++ b/tests/test_processing.py
@@ -93,6 +93,20 @@ def test_is_nan():
         is_nan(pl.Series([np.nan, 1.0, None])).to_numpy(),
         np.array([True, False, None]),
     )
+    for series in (
+        pl.Series(["a", None, "b"]),
+        pl.Series([True, None, False]),
+        pl.Series([1, None, 2]),
+        pl.Series(["a", None, "b"], dtype=pl.Categorical),
+    ):
+        np.testing.assert_equal(
+            is_nan(series).to_numpy(),
+            np.array([False, False, False]),
+        )
+        np.testing.assert_equal(
+            is_nan_or_none(series).to_numpy(),
+            np.array([False, True, False]),
+        )
 
 
 def test_is_none():
diff --git a/utilsforecast/processing.py b/utilsforecast/processing.py
index c6bf34d..691fcfd 100644
--- a/utilsforecast/processing.py
+++ b/utilsforecast/processing.py
@@ -233,8 +233,11 @@ def filter_with_mask(
 def is_nan(s: Series) -> Series:
     if isinstance(s, pd.Series):
         out = s.isna()
-    else:
+    elif s.dtype.is_float():
         out = s.is_nan()
+    else:
+        # polars only supports is_nan on numeric columns and only floats can hold NaN
+        out = pl.repeat(False, s.len(), eager=True).alias(s.name)
     return out