diff --git a/daft/table/table.py b/daft/table/table.py
index 054ba099f9..eb39859814 100644
--- a/daft/table/table.py
+++ b/daft/table/table.py
@@ -475,8 +475,6 @@ def read_parquet_into_pyarrow(
         multithreaded_io=multithreaded_io,
         coerce_int96_timestamp_unit=coerce_int96_timestamp_unit._timeunit,
     )
-    if len(metadata) == 0:
-        metadata = None
     schema = pa.schema(fields, metadata=metadata)
     columns = [pa.chunked_array(c) for c in columns]  # type: ignore
     return pa.table(columns, schema=schema)
diff --git a/tests/integration/io/parquet/test_reads_public_data.py b/tests/integration/io/parquet/test_reads_public_data.py
index 116e675b57..c7aa75ded7 100644
--- a/tests/integration/io/parquet/test_reads_public_data.py
+++ b/tests/integration/io/parquet/test_reads_public_data.py
@@ -232,7 +232,7 @@ def test_parquet_read_table_into_pyarrow(parquet_file, public_storage_io_config,
     )
     pa_read = read_parquet_with_pyarrow(url)
     assert daft_native_read.schema == pa_read.schema
-    assert daft_native_read.schema.metadata == pa_read.schema.metadata
+    assert pa_read.schema.metadata is None or daft_native_read.schema.metadata == pa_read.schema.metadata
     pd.testing.assert_frame_equal(daft_native_read.to_pandas(), pa_read.to_pandas())