diff --git a/python/paddle/fluid/initializer.py b/python/paddle/fluid/initializer.py
index c838f09ced26e1f21f7a7c9cfed5095323b1cead..6c394f4c6955fdccae79fa7bca791f00bba41d38 100644
--- a/python/paddle/fluid/initializer.py
+++ b/python/paddle/fluid/initializer.py
@@ -783,7 +783,7 @@ class BilinearInitializer(Initializer):
         weight = np.reshape(weight, shape)
 
         # to be compatible of fp16 initalizers
-        if var.dtype == VarDesc.VarType.FP16:
+        if var.dtype == VarDesc.VarType.FP16 or var.dtype == VarDesc.VarType.FP64:
             out_dtype = VarDesc.VarType.FP32
             out_var = block.create_var(
                 name=unique_name.generate(".".join(
@@ -800,7 +800,8 @@ class BilinearInitializer(Initializer):
             value_name = "fp32_values"
             values = [float(v) for v in weight.flat]
         else:
-            raise ValueError("Unsupported dtype %s", input.dtype)
+            raise TypeError("Unsupported dtype %s", var.dtype)
+
         if np.prod(shape) > 1024 * 1024:
             raise ValueError("The size of input is too big. ")
         op = block.append_op(
@@ -812,7 +813,7 @@ class BilinearInitializer(Initializer):
                 value_name: values
             })
 
-        if var.dtype == VarDesc.VarType.FP16:
+        if var.dtype == VarDesc.VarType.FP16 or var.dtype == VarDesc.VarType.FP64:
             block.append_op(
                 type="cast",
                 inputs={"X": out_var},
diff --git a/python/paddle/fluid/tests/unittests/test_initializer.py b/python/paddle/fluid/tests/unittests/test_initializer.py
index c6bed4db72e50135fba7b22f805efb281c178e2d..1b880569dfee0b7a20e783fae419e32db712121e 100644
--- a/python/paddle/fluid/tests/unittests/test_initializer.py
+++ b/python/paddle/fluid/tests/unittests/test_initializer.py
@@ -474,18 +474,24 @@ class TestBilinearInitializer(unittest.TestCase):
                 lod_level=0,
                 name="param",
                 initializer=initializer.BilinearInitializer())
-        num_ops = 2 if dtype == "float16" else 1
+        num_ops = 2 if dtype == "float16" or dtype == "float64" else 1
         self.assertEqual(len(block.ops), num_ops)
         init_op = block.ops[0]
         self.assertEqual(init_op.type, 'assign_value')
         return block
 
+    def test_bilinear_initializer_fp64(self):
+        self.test_bilinear_initializer(dtype='float64')
+
     def test_bilinear_initializer_fp16(self):
         """Test the bilinear initializer with supplied arguments
         """
         block = self.test_bilinear_initializer("float16")
         self.assertTrue(check_cast_op(block.ops[1]))
 
+    def test_type_error(self):
+        self.assertRaises(TypeError, self.test_bilinear_initializer, 'int32')
+
 
 class TestNumpyArrayInitializer(unittest.TestCase):
     def test_numpy_array_initializer(self, dtype="float32"):