diff --git a/pyopencl/tools.py b/pyopencl/tools.py
index 2a2d0f9d061d6f74c38dc07a4341fbf8081d463d..633854e07a523b2cac694b111fabb4b7422ad73c 100644
--- a/pyopencl/tools.py
+++ b/pyopencl/tools.py
@@ -249,11 +249,18 @@ def pytest_generate_tests_for_pyopencl(metafunc):
                     (self.device.name.strip(),
                      self.device.platform.name.strip()))
 
+    class QueueFactory:
+        def __init__(self, ctx_factory):
+            self.ctx_factory = ctx_factory
+
+        def __call__(self):
+            return cl.CommandQueue(self.ctx_factory())
+
     test_plat_and_dev = get_test_platforms_and_devices()
 
     arg_names = []
 
-    for arg in ("platform", "device", "ctx_factory", "ctx_getter"):
+    for arg in ("platform", "device", "ctx_factory", "cl_queue", "ctx_getter"):
         if arg not in metafunc.fixturenames:
             continue
 
@@ -277,6 +284,7 @@ def pytest_generate_tests_for_pyopencl(metafunc):
         for device in plat_devs:
             arg_dict["device"] = device
             arg_dict["ctx_factory"] = ContextFactory(device)
+            arg_dict["cl_queue"] = QueueFactory(ContextFactory(device))
             arg_dict["ctx_getter"] = ContextFactory(device)
 
             arg_values.append(tuple(arg_dict[name] for name in arg_names))
diff --git a/test/test_wrapper.py b/test/test_wrapper.py
index 75b39b495feab12b18f89b99d3e85bd53ea08849..74376b35fefedd5d6a795c6436347a843180e81b 100644
--- a/test/test_wrapper.py
+++ b/test/test_wrapper.py
@@ -531,6 +531,24 @@ def test_copy_buffer(ctx_factory):
     assert la.norm(a - b) == 0
 
 
+def test_copy_buffer_using_cl_queue(ctx_factory, cl_queue):
+    context = ctx_factory()
+
+    queue = cl_queue()
+    mf = cl.mem_flags
+
+    a = np.random.rand(50000).astype(np.float32)
+    b = np.empty_like(a)
+
+    buf1 = cl.Buffer(context, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=a)
+    buf2 = cl.Buffer(context, mf.WRITE_ONLY, b.nbytes)
+
+    cl.enqueue_copy(queue, buf2, buf1).wait()
+    cl.enqueue_copy(queue, b, buf2).wait()
+
+    assert la.norm(a - b) == 0
+
+
 def test_mempool(ctx_factory):
     from pyopencl.tools import MemoryPool, ImmediateAllocator