diff --git a/flink-python/pyflink/datastream/stream_execution_environment.py b/flink-python/pyflink/datastream/stream_execution_environment.py
index c69e105347aa3..aabd203175eac 100644
--- a/flink-python/pyflink/datastream/stream_execution_environment.py
+++ b/flink-python/pyflink/datastream/stream_execution_environment.py
@@ -767,6 +767,18 @@ def from_source(self,
             j_type_info = type_info.get_java_type_info()
         else:
             j_type_info = None
+        if watermark_strategy._timestamp_assigner is not None:
+            # in case users have specified a custom Python TimestampAssigner, the timestamp
+            # extraction and watermark generation cannot be performed on the Java source. So we
+            # create the source without watermarks first and then apply the watermark strategy
+            # downstream, where the custom Python TimestampAssigner can be executed.
+            j_data_stream = self._j_stream_execution_environment.fromSource(
+                source.get_java_function(),
+                WatermarkStrategy.no_watermarks()._j_watermark_strategy,
+                source_name,
+                j_type_info)
+            return DataStream(j_data_stream=j_data_stream).assign_timestamps_and_watermarks(
+                watermark_strategy)
         j_data_stream = self._j_stream_execution_environment.fromSource(
             source.get_java_function(),
             watermark_strategy._j_watermark_strategy,
diff --git a/flink-python/pyflink/datastream/tests/test_stream_execution_environment.py b/flink-python/pyflink/datastream/tests/test_stream_execution_environment.py
index 8a10244a8a3bd..61f3a4411c5f8 100644
--- a/flink-python/pyflink/datastream/tests/test_stream_execution_environment.py
+++ b/flink-python/pyflink/datastream/tests/test_stream_execution_environment.py
@@ -27,12 +27,14 @@
 
 from pyflink.common import Configuration, ExecutionConfig
 from pyflink.common.typeinfo import Types
+from pyflink.common.watermark_strategy import WatermarkStrategy, TimestampAssigner
 from pyflink.datastream import (StreamExecutionEnvironment, CheckpointConfig,
                                 CheckpointingMode, SlotSharingGroup)
 from pyflink.datastream.connectors.kafka import FlinkKafkaConsumer
+from pyflink.datastream.connectors.number_seq import NumberSequenceSource
 from pyflink.datastream.execution_mode import RuntimeExecutionMode
 from pyflink.datastream.formats.json import JsonRowDeserializationSchema
-from pyflink.datastream.functions import SourceFunction
+from pyflink.datastream.functions import ProcessFunction, SourceFunction
 from pyflink.datastream.slot_sharing_group import MemorySize
 from pyflink.datastream.tests.test_util import DataStreamTestSinkFunction
 from pyflink.find_flink_home import _find_flink_source_root
@@ -674,5 +676,35 @@ def test_register_cached_file(self):
         self.assertEqual(cached_files.size(), 1)
         self.assertEqual(cached_files[0].getField(0), 'cache_test')
 
+    def test_from_source_with_timestamp_assigner(self):
+        self.env.set_parallelism(1)
+        self.env.get_config().set_auto_watermark_interval(2000)
+        seq_source = NumberSequenceSource(1, 4)
+
+        class MyTimestampAssigner(TimestampAssigner):
+
+            def extract_timestamp(self, value, record_timestamp) -> int:
+                return value * 1000
+
+        class MyProcessFunction(ProcessFunction):
+
+            def process_element(self, value, ctx):
+                yield "value: {}, timestamp: {}".format(str(value), str(ctx.timestamp()))
+
+        watermark_strategy = WatermarkStrategy.for_monotonous_timestamps() \
+            .with_timestamp_assigner(MyTimestampAssigner())
+        ds = self.env.from_source(seq_source, watermark_strategy, "seq_source",
+                                  type_info=Types.LONG())
+        ds.process(MyProcessFunction(), output_type=Types.STRING()).add_sink(self.test_sink)
+        self.env.execute('test from source with timestamp assigner')
+        results = self.test_sink.get_results()
+        expected = ["value: 1, timestamp: 1000",
+                    "value: 2, timestamp: 2000",
+                    "value: 3, timestamp: 3000",
+                    "value: 4, timestamp: 4000"]
+        results.sort()
+        expected.sort()
+        self.assertEqual(expected, results)
+
     def tearDown(self) -> None:
         self.test_sink.clear()