diff --git a/src/spatialdata/_core/spatialdata.py b/src/spatialdata/_core/spatialdata.py
index 8be3c2682..9f5c91c7a 100644
--- a/src/spatialdata/_core/spatialdata.py
+++ b/src/spatialdata/_core/spatialdata.py
@@ -215,9 +215,9 @@ def validate_table_in_spatialdata(self, table: AnnData) -> None:
                     else:
                         dtype = element.index.dtype
                     instance_dtype = self.get_instance_key_column(table).dtype
-                    if dtype != instance_dtype and (
-                        pd.api.types.is_string_dtype(dtype) or pd.api.types.is_string_dtype(instance_dtype)
-                    ):
+                    # Only a string vs non-string mismatch is an error: string ids may be stored as `object` or
+                    # as a pandas string dtype (the default with pandas>=3) depending on the reader.
+                    if pd.api.types.is_string_dtype(dtype) != pd.api.types.is_string_dtype(instance_dtype):
                         raise TypeError(
                             f"Table instance_key column ({instance_key}) has a dtype "
                             f"({instance_dtype}) that does not match the dtype of the indices of "
diff --git a/tests/datasets/test_datasets.py b/tests/datasets/test_datasets.py
index 3fc66370c..d71afedd2 100644
--- a/tests/datasets/test_datasets.py
+++ b/tests/datasets/test_datasets.py
@@ -2,10 +2,11 @@
 
 from pathlib import Path
 
+import pandas as pd
 import pooch
 import pytest
 
-from spatialdata import SpatialData
+from spatialdata import SpatialData, read_zarr
 from spatialdata.datasets import _cache_dir, _shipped_registry, blobs, cells, raccoon
 
 
@@ -70,3 +71,23 @@ def test_cells_download(tmp_path) -> None:
     assert len(sdata.shapes["nucleus_boundaries"]) == 94
     assert len(sdata.points["transcripts"].compute()) == 19479
     assert sdata.tables["table"].shape == (94, 5101)
+
+
+@pytest.mark.network
+def test_cells_string_instance_key_dtype(tmp_path) -> None:
+    # Regression: the `cell_boundaries` index is read as the pandas>=3 `str` dtype while the table's `cell_id`
+    # column is read as `object`; both hold string ids, so annotating by `cell_id` must validate and round-trip.
+    sdata = cells(path=str(tmp_path / "cache"))
+    table = sdata.tables["table"]
+    table.obs["region"] = pd.Categorical(["cell_boundaries"] * table.n_obs)
+    sdata.set_table_annotates_spatialelement("table", region="cell_boundaries", instance_key="cell_id")
+    sdata.validate_table_in_spatialdata(table)
+
+    sdata.write(tmp_path / "data.zarr")
+    sdata_read = read_zarr(tmp_path / "data.zarr")
+    assert sdata_read.tables["table"].uns["spatialdata_attrs"]["instance_key"] == "cell_id"
+
+    # a string vs integer mismatch is still an error
+    sdata.set_table_annotates_spatialelement("table", region="cell_boundaries", instance_key="cell_labels")
+    with pytest.raises(TypeError, match="does not match the dtype"):
+        sdata.validate_table_in_spatialdata(table)