diff --git a/pyiceberg/table/__init__.py b/pyiceberg/table/__init__.py
index df7aaf824f..48cd34eb31 100644
--- a/pyiceberg/table/__init__.py
+++ b/pyiceberg/table/__init__.py
@@ -275,6 +275,8 @@ def __init__(self, table: Table, autocommit: bool = False):
 
     @property
     def table_metadata(self) -> TableMetadata:
+        if not self._updates:
+            return self._table.metadata
         return update_table_metadata(self._table.metadata, self._updates)
 
     def __enter__(self) -> Transaction:
diff --git a/tests/table/test_init.py b/tests/table/test_init.py
index 481326b44b..384704324f 100644
--- a/tests/table/test_init.py
+++ b/tests/table/test_init.py
@@ -1103,6 +1103,20 @@ def test_update_metadata_table_schema(table_v2: Table) -> None:
     assert new_metadata.current_schema_id == 2
 
 
+def test_transaction_table_metadata_without_staged_updates(table_v2: Table) -> None:
+    transaction = table_v2.transaction()
+    assert transaction.table_metadata is table_v2.metadata
+
+
+def test_transaction_table_metadata_with_staged_updates(table_v2: Table) -> None:
+    transaction = table_v2.transaction()
+    transaction.set_properties(test_a="test_a")
+
+    assert transaction.table_metadata.properties["test_a"] == "test_a"
+    assert transaction.table_metadata is not table_v2.metadata
+    assert "test_a" not in table_v2.metadata.properties
+
+
 def test_update_metadata_add_snapshot(table_v2: Table) -> None:
     new_snapshot = Snapshot(
         snapshot_id=25,
diff --git a/tests/table/test_snapshots.py b/tests/table/test_snapshots.py
index 39aeb7b349..f169dafe7a 100644
--- a/tests/table/test_snapshots.py
+++ b/tests/table/test_snapshots.py
@@ -674,6 +674,8 @@ def make_file() -> DataFile:
         return DataFile.from_args(content=DataFileContent.DATA, record_count=1, file_size_in_bytes=1, partition=Record())
 
     txn = table_v2.transaction()
+    # Stage an update, so every read calls update_table_metadata
+    txn.set_properties(test_a="test_a")
 
     with mock.patch("pyiceberg.table.update_table_metadata", wraps=update_table_metadata) as spy:
         # _summary() cost must not scale with the number of data files