diff --git a/src/hdmf_zarr/backend.py b/src/hdmf_zarr/backend.py
index fa8cab43..fbdf91f3 100644
--- a/src/hdmf_zarr/backend.py
+++ b/src/hdmf_zarr/backend.py
@@ -1033,7 +1033,8 @@ def write_dataset(self, **kwargs):  # noqa: C901
                     if field['dtype'] is str or field['dtype'] in (
                             'str', 'text', 'utf', 'utf8', 'utf-8', 'isodatetime'
                     ):
-                        new_dtype.append((field['name'], 'U25'))
+                        # Zarr does not support variable length strings
+                        new_dtype.append((field['name'], 'O'))
                     elif isinstance(field['dtype'], dict):
                         # eg. for some references, dtype will be of the form
                         # {'target_type': 'Baz', 'reftype': 'object'}
diff --git a/tests/unit/base_tests_zarrio.py b/tests/unit/base_tests_zarrio.py
index 81ab9416..ec1758cd 100644
--- a/tests/unit/base_tests_zarrio.py
+++ b/tests/unit/base_tests_zarrio.py
@@ -436,7 +436,7 @@ def test_read_reference_compound(self):
         read_builder = self.root['ref_dataset']
 
         # ensure the array was written as a compound array
-        ref_dtype = np.dtype([('id', '<i4'), ('name', '<U25'), ('reference', 'O')])
+        ref_dtype = np.dtype([('id', '<i4'), ('name', 'O'), ('reference', 'O')])
         self.assertEqual(read_builder.data.dataset.dtype, ref_dtype)
 
         # Load the elements of each entry in the compound dataset and compar the index, string, and referenced array