diff --git a/biopandas/pdb/pandas_pdb.py b/biopandas/pdb/pandas_pdb.py
index 6ed68cb..c4d2a00 100644
--- a/biopandas/pdb/pandas_pdb.py
+++ b/biopandas/pdb/pandas_pdb.py
@@ -24,7 +24,7 @@
 
 from biopandas.constants import ATOMIC_MASSES
 
-from .engines import amino3to1dict, pdb_df_columns, pdb_records
+from .engines import amino3to1dict, pdb_records
 
 pd_version = LooseVersion(pd.__version__)
 
@@ -823,9 +823,11 @@ def to_pdb(self, path, records=None, gz=False, append_newline=True):
                     for idx in range(dfs[r][c].values.shape[0]):
                         if len(dfs[r][c].values[idx]) > 8:
                             dfs[r][c].values[idx] = str(dfs[r][c].values[idx]).strip()
-                if c in {"line_idx", "OUT"}:
+                if c in {"line_idx", "OUT", "model_id"}:
                     pass
-                elif r in {"ATOM", "HETATM"} and c not in pdb_df_columns:
+                elif r in {"ATOM", "HETATM", "ANISOU"} and c not in {
+                    col["id"] for col in pdb_records[r]
+                }:
                     warn(
                         "Column %s is not an expected column and"
                         " will be skipped." % c
diff --git a/docs/CHANGELOG.md b/docs/CHANGELOG.md
index 6c0d62f..0e47d91 100755
--- a/docs/CHANGELOG.md
+++ b/docs/CHANGELOG.md
@@ -11,6 +11,8 @@ The CHANGELOG for the current development version is available at
 
 - Fix: `PandasMol2.rmsd()` now normalizes by the number of atoms it actually compared. With the default `heavy_only=True`, the sum of squared deviations over the hydrogen-filtered atoms was divided by the total atom count including the discarded hydrogens, silently understating the result by a factor of `sqrt(n_total / n_heavy)`. Addresses issue [#163](https://github.com/BioPandas/biopandas/issues/163).
 
+- Fix: `PandasPdb.to_pdb()` no longer fails with a `TypeError` on structures with `ANISOU` records after `label_models()`. The `model_id` column is now skipped silently, like in `to_pdb_stream()`, and other unexpected `ANISOU` columns are skipped with a warning, as for `ATOM` and `HETATM`. Addresses issue [#124](https://github.com/BioPandas/biopandas/issues/124).
+
 ### 0.5.2 (24/08/2026)
 
 - Fix: improves robustness of `amino3to1` residue identification by handling `NaN` insertion codes (NumPy 2.4+ compatibility) and including chain IDs and separators to avoid collisions. PR [#157](https://github.com/BioPandas/biopandas/pull/157)
diff --git a/tests/pdb/test_write_pdb.py b/tests/pdb/test_write_pdb.py
index 0b183b7..5f4b48b 100644
--- a/tests/pdb/test_write_pdb.py
+++ b/tests/pdb/test_write_pdb.py
@@ -84,6 +84,35 @@ def test_anisou():
     assert f1 == four_eiy
 
 
+def test_anisou_with_model_id():
+    """Test writing ANISOU entries after labelling models."""
+    ppdb = PandasPdb()
+    ppdb.read_pdb(TESTDATA_FILENAME2)
+    ppdb.label_models()
+    assert "model_id" in ppdb.df["ANISOU"].columns
+    with warnings.catch_warnings(record=True) as w:
+        warnings.simplefilter("always")
+        ppdb.to_pdb(path=OUTFILE, records=None)
+    assert not any("model_id" in str(x.message) for x in w)
+    with open(OUTFILE, "r") as f:
+        f1 = f.read()
+    os.remove(OUTFILE)
+    assert f1 == four_eiy
+
+
+def test_anisou_nonexpected_column():
+    """Test that unexpected ANISOU columns are skipped with a warning."""
+    ppdb = PandasPdb()
+    ppdb.read_pdb(TESTDATA_FILENAME2)
+    ppdb.df["ANISOU"]["test"] = 1
+    with pytest.warns(UserWarning, match="Column test is not an expected"):
+        ppdb.to_pdb(path=OUTFILE, records=None)
+    with open(OUTFILE, "r") as f:
+        f1 = f.read()
+    os.remove(OUTFILE)
+    assert f1 == four_eiy
+
+
 def test_add_remark():
     """Test adding a REMARK entry."""
     # Add remark