diff --git a/keras2onnx/ke2onnx/bidirectional.py b/keras2onnx/ke2onnx/bidirectional.py
index f4cbc934..bc2eee2d 100644
--- a/keras2onnx/ke2onnx/bidirectional.py
+++ b/keras2onnx/ke2onnx/bidirectional.py
@@ -240,10 +240,17 @@ def convert_bidirectional(scope, operator, container):
             apply_split(scope, transposed_y_name, [forward_y_name, backward_y_name], container, axis=2)
 
             # Change (T, N, 1, C') into (T, N, C') to meet Keras spec
-            container.add_node('Squeeze', forward_y_name, operator.outputs[0].full_name,
+            forward_y_name_1 = scope.get_unique_variable_name(operator.full_name + '_Y_forward_1')
+            backward_y_name_1 = scope.get_unique_variable_name(operator.full_name + '_Y_backward_1')
+            container.add_node('Squeeze', forward_y_name, forward_y_name_1,
                                name=scope.get_unique_variable_name('Squeeze'), axes=[2])
-            container.add_node('Squeeze', backward_y_name, operator.outputs[1].full_name,
+            container.add_node('Squeeze', backward_y_name, backward_y_name_1,
                                name=scope.get_unique_variable_name('Squeeze'), axes=[2])
+
+            apply_reshape(scope, forward_y_name_1, operator.outputs[0].full_name, container,
+                          desired_shape=[-1, seq_length, hidden_size])
+            apply_reshape(scope, backward_y_name_1, operator.outputs[1].full_name, container,
+                          desired_shape=[-1, seq_length, hidden_size])
     else:
         perm = [1, 0, 2]
         if merge_concat:
diff --git a/tests/test_layers.py b/tests/test_layers.py
index 45df6324..ddd5a052 100644
--- a/tests/test_layers.py
+++ b/tests/test_layers.py
@@ -642,19 +642,34 @@ def test_LSTM_reshape(self):
         self.assertTrue(self.run_onnx_runtime('tf_lstm', onnx_model, data, expected))
 
     def test_Bidirectional(self):
-        input_dim = 10
-        sequence_len = 5
-        model = keras.Sequential()
-        model.add(keras.layers.Bidirectional(keras.layers.LSTM(10, return_sequences=False),
-                  input_shape=(5, 10)))
-        model.add(keras.layers.Dense(5))
-        model.add(keras.layers.Activation('softmax'))
-        model.compile(loss='categorical_crossentropy', optimizer='rmsprop')
-
-        onnx_model = keras2onnx.convert_keras(model, 'test')
-        data = np.random.rand(input_dim, sequence_len).astype(np.float32).reshape((1, sequence_len, input_dim))
-        expected = model.predict(data)
-        self.assertTrue(self.run_onnx_runtime('bidirectional', onnx_model, data, expected))
+        for return_sequences in [True, False]:
+            input_dim = 10
+            sequence_len = 5
+            model = keras.Sequential()
+            model.add(keras.layers.Bidirectional(keras.layers.LSTM(10, return_sequences=return_sequences),
+                      input_shape=(5, 10)))
+            model.add(keras.layers.Dense(5))
+            model.add(keras.layers.Activation('softmax'))
+            model.compile(loss='categorical_crossentropy', optimizer='rmsprop')
+
+            onnx_model = keras2onnx.convert_keras(model, 'test')
+            data = np.random.rand(input_dim, sequence_len).astype(np.float32).reshape((1, sequence_len, input_dim))
+            expected = model.predict(data)
+            self.assertTrue(self.run_onnx_runtime('bidirectional', onnx_model, data, expected))
+
+        for merge_mode in ['concat', None]:
+            # TODO: case return_sequences=False
+            for return_sequences in [True]:
+                input_dim = 10
+                sequence_len = 5
+                sub_input1 = keras.layers.Input(shape=(sequence_len, input_dim))
+                sub_mapped1 = keras.layers.Bidirectional(keras.layers.LSTM(10, return_sequences=return_sequences),
+                                                     input_shape=(5, 10), merge_mode=merge_mode)(sub_input1)
+                keras_model = keras.Model(inputs=sub_input1, outputs=sub_mapped1)
+                onnx_model = keras2onnx.convert_keras(keras_model, 'test_2')
+                data = np.random.rand(input_dim, sequence_len).astype(np.float32).reshape((1, sequence_len, input_dim))
+                expected = keras_model.predict(data)
+                self.assertTrue(self.run_onnx_runtime('bidirectional', onnx_model, data, expected))
 
     def test_separable_convolution(self):
         N, C, H, W = 2, 3, 5, 5