diff --git a/chainer/backend.py b/chainer/backend.py
index 0be00d379048..1540ecb001a1 100644
--- a/chainer/backend.py
+++ b/chainer/backend.py
@@ -67,7 +67,8 @@ def copyto(dst, src):
     elif isinstance(dst, cuda.ndarray):
         if isinstance(src, chainer.get_cpu_array_types()):
             src = numpy.asarray(src)
-            if dst.flags.c_contiguous or dst.flags.f_contiguous:
+            if (src.dtype == dst.dtype
+                    and (dst.flags.c_contiguous or dst.flags.f_contiguous)):
                 dst.set(src)
             else:
                 cuda.cupy.copyto(dst, cuda.to_gpu(src, device=dst.device))
diff --git a/tests/chainer_tests/test_backend.py b/tests/chainer_tests/test_backend.py
index 1d6c19620206..330eaf636f3b 100644
--- a/tests/chainer_tests/test_backend.py
+++ b/tests/chainer_tests/test_backend.py
@@ -65,15 +65,21 @@ def test_from_chx_cuda(self):
         numpy.testing.assert_array_equal(self._to_cpu(dst), self.src_data)
 
 
+@testing.parameterize(*testing.product({
+    'dtype': [numpy.float16, numpy.float32],
+}))
 class TestCopyToCPU(_TestCopyToBase, unittest.TestCase):
     def _get_dst(self):
-        return self.dst_data
+        return self.dst_data.astype(self.dtype, copy=False)
 
 
+@testing.parameterize(*testing.product({
+    'dtype': [numpy.float16, numpy.float32],
+}))
 @attr.gpu
 class TestCopyToGPU(_TestCopyToBase, unittest.TestCase):
     def _get_dst(self):
-        return cuda.cupy.array(self.dst_data)
+        return cuda.cupy.array(self.dst_data, self.dtype)
 
     @attr.multi_gpu(2)
     def test_gpu_to_another_gpu(self):
@@ -92,17 +98,23 @@ def _get_dst(self):
         return dst
 
 
+@testing.parameterize(*testing.product({
+    'dtype': [numpy.float16, numpy.float32],
+}))
 @attr.chainerx
 class TestCopyToChxNative(_TestCopyToBase, unittest.TestCase):
     def _get_dst(self):
-        return chainerx.array(self.dst_data, device='native')
+        return chainerx.array(self.dst_data, dtype=self.dtype, device='native')
 
 
+@testing.parameterize(*testing.product({
+    'dtype': [numpy.float16, numpy.float32],
+}))
 @attr.chainerx
 @attr.gpu
 class TestCopyToChxCuda(_TestCopyToBase, unittest.TestCase):
     def _get_dst(self):
-        return chainerx.array(self.dst_data, device='cuda:0')
+        return chainerx.array(self.dst_data, dtype=self.dtype, device='cuda:0')
 
 
 class TestCopyToError(unittest.TestCase):