diff --git a/include/uv.h b/include/uv.h
index 483a174282..bcb9985066 100644
--- a/include/uv.h
+++ b/include/uv.h
@@ -1372,6 +1372,13 @@ UV_EXTERN int uv_work_queue(uv_loop_t* loop,
                             uv_work_cb work_cb,
                             uv_after_work_cb after_work_cb);
 
+/* Cancel a pending work request. Fails if the work request is executing or
+ * has finished executing.
+ *
+ * Returns 0 on success, -1 on error. The loop error code is not touched.
+ */
+UV_EXTERN int uv_work_cancel(uv_work_t* req);
+
 
 struct uv_cpu_info_s {
   char* model;
diff --git a/src/unix/threadpool.c b/src/unix/threadpool.c
index 68b2a78980..243768ddfd 100644
--- a/src/unix/threadpool.c
+++ b/src/unix/threadpool.c
@@ -30,6 +30,9 @@ static ngx_queue_t wq = { &wq, &wq };
 static volatile int initialized;
 
 
+/* To avoid deadlock with uv_work_cancel() it's crucial that the worker
+ * never holds the global mutex and the loop-local mutex at the same time.
+ */
 static void worker(void* arg) {
   struct uv__work* w;
   ngx_queue_t* q;
@@ -46,8 +49,11 @@ static void worker(void* arg) {
 
     if (q == &exit_message)
       uv_cond_signal(&cond);
-    else
+    else {
       ngx_queue_remove(q);
+      ngx_queue_init(q);  /* Signal uv_work_cancel() that the work req is
+                             executing. */
+    }
 
     uv_mutex_unlock(&mutex);
 
@@ -58,6 +64,8 @@ static void worker(void* arg) {
     w->work(w);
 
     uv_mutex_lock(&w->loop->wq_mutex);
+    w->work = NULL;  /* Signal uv_work_cancel() that the work req is done
+                        executing. */
     ngx_queue_insert_tail(&w->loop->wq, &w->wq);
     uv_async_send(&w->loop->wq_async);
     uv_mutex_unlock(&w->loop->wq_mutex);
@@ -176,6 +184,30 @@ int uv_work_queue(uv_loop_t* loop,
 }
 
 
+int uv_work_cancel(uv_work_t* req) {
+  struct uv__work* w;
+  int cancelled;
+
+  w = &req->work_req;
+  uv_mutex_lock(&mutex);
+  uv_mutex_lock(&w->loop->wq_mutex);
+
+  cancelled = !ngx_queue_empty(&w->wq) && w->work != NULL;
+  if (cancelled)
+    ngx_queue_remove(&w->wq);
+
+  uv_mutex_unlock(&w->loop->wq_mutex);
+  uv_mutex_unlock(&mutex);
+
+  if (cancelled) {
+    uv__req_unregister(req->loop, req);
+    return 0;
+  }
+
+  return -1;
+}
+
+
 int uv_queue_work(uv_loop_t* loop,
                   uv_work_t* req,
                   uv_work_cb work_cb,
diff --git a/src/win/threadpool.c b/src/win/threadpool.c
index 8278aef46b..37928b9447 100644
--- a/src/win/threadpool.c
+++ b/src/win/threadpool.c
@@ -70,6 +70,11 @@ int uv_work_queue(uv_loop_t* loop, uv_work_t* req, uv_work_cb work_cb,
 }
 
 
+int uv_work_cancel(uv_work_t* req) {
+  return -1;
+}
+
+
 void uv_process_work_req(uv_loop_t* loop, uv_work_t* req) {
   uv__req_unregister(loop, req);
   if(req->after_work_cb)
diff --git a/test/test-list.h b/test/test-list.h
index f5cc95c6b3..3ae4ecc058 100644
--- a/test/test-list.h
+++ b/test/test-list.h
@@ -186,6 +186,7 @@ TEST_DECLARE   (fs_rename_to_existing_file)
 TEST_DECLARE   (threadpool_queue_work_simple)
 TEST_DECLARE   (threadpool_queue_work_einval)
 TEST_DECLARE   (threadpool_multiple_event_loops)
+TEST_DECLARE   (threadpool_cancel_work)
 TEST_DECLARE   (thread_mutex)
 TEST_DECLARE   (thread_rwlock)
 TEST_DECLARE   (thread_create)
@@ -451,6 +452,7 @@ TASK_LIST_START
   TEST_ENTRY  (threadpool_queue_work_simple)
   TEST_ENTRY  (threadpool_queue_work_einval)
   TEST_ENTRY  (threadpool_multiple_event_loops)
+  TEST_ENTRY  (threadpool_cancel_work)
   TEST_ENTRY  (thread_mutex)
   TEST_ENTRY  (thread_rwlock)
   TEST_ENTRY  (thread_create)
diff --git a/test/test-threadpool-cancel.c b/test/test-threadpool-cancel.c
new file mode 100644
index 0000000000..d6f7f77618
--- /dev/null
+++ b/test/test-threadpool-cancel.c
@@ -0,0 +1,113 @@
+/* Copyright Joyent, Inc. and other Node contributors. All rights reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ */
+
+#include "uv.h"
+#include "task.h"
+
+static uv_cond_t signal_cond;
+static uv_mutex_t signal_mutex;
+static uv_mutex_t wait_mutex;
+static unsigned done_cb_called;
+static unsigned timer_cb_called;
+static uv_work_t reqs[16];
+
+
+static void work_cb(uv_work_t* req) {
+  uv_mutex_lock(&signal_mutex);
+  uv_cond_signal(&signal_cond);
+  uv_mutex_unlock(&signal_mutex);
+
+  uv_mutex_lock(&wait_mutex);
+  uv_mutex_unlock(&wait_mutex);
+}
+
+
+static void done_cb(uv_work_t* req) {
+  done_cb_called++;
+  free(req);
+}
+
+
+static void fail_cb(uv_work_t* req) {
+  ASSERT(0 && "fail_cb called");
+}
+
+
+static void timer_cb(uv_timer_t* handle, int status) {
+  unsigned i;
+
+  for (i = 0; i < ARRAY_SIZE(reqs); i++)
+    ASSERT(0 == uv_work_cancel(reqs + i));
+
+  uv_close((uv_handle_t*) handle, NULL);
+  uv_mutex_unlock(&signal_mutex);
+  uv_mutex_unlock(&wait_mutex);
+  timer_cb_called++;
+}
+
+
+/* Submit as many work requests as it takes to saturate the thread pool,
+ * then submit more and try to cancel them.
+ */
+TEST_IMPL(threadpool_cancel_work) {
+  uv_timer_t timer_handle;
+  unsigned num_threads;
+  unsigned i;
+  uv_work_t* req;
+
+  ASSERT(0 == uv_cond_init(&signal_cond));
+  ASSERT(0 == uv_mutex_init(&signal_mutex));
+  ASSERT(0 == uv_mutex_init(&wait_mutex));
+
+  uv_mutex_lock(&signal_mutex);
+  uv_mutex_lock(&wait_mutex);
+
+  for (num_threads = 0; /* empty */; num_threads++) {
+    req = malloc(sizeof(*req));
+    ASSERT(req != NULL);
+    ASSERT(0 == uv_work_queue(uv_default_loop(), req, work_cb, done_cb));
+
+    /* Expect to get signalled within 350 ms, otherwise assume that
+     * the thread pool is saturated. As with any timing dependent test,
+     * this is obviously not ideal.
+     */
+    if (uv_cond_timedwait(&signal_cond, &signal_mutex, 350 * 1e6)) {
+      ASSERT(0 == uv_work_cancel(req));
+      free(req);
+      break;
+    }
+  }
+
+  for (i = 0; i < ARRAY_SIZE(reqs); i++)
+    ASSERT(0 == uv_work_queue(uv_default_loop(), reqs + i, fail_cb, NULL));
+
+  ASSERT(0 == uv_timer_init(uv_default_loop(), &timer_handle));
+  ASSERT(0 == uv_timer_start(&timer_handle, timer_cb, 10, 0));
+  ASSERT(0 == uv_run(uv_default_loop()));
+
+  ASSERT(done_cb_called == num_threads);
+
+  uv_cond_destroy(&signal_cond);
+  uv_mutex_destroy(&signal_mutex);
+  uv_mutex_destroy(&wait_mutex);
+
+  return 0;
+}
diff --git a/uv.gyp b/uv.gyp
index 749decdea7..c1cff46caf 100644
--- a/uv.gyp
+++ b/uv.gyp
@@ -299,6 +299,7 @@
         'test/test-tcp-writealot.c',
         'test/test-tcp-unexpected-read.c',
         'test/test-threadpool.c',
+        'test/test-threadpool-cancel.c',
         'test/test-mutexes.c',
         'test/test-thread.c',
         'test/test-barrier.c',