diff --git a/tokio-threadpool/Cargo.toml b/tokio-threadpool/Cargo.toml
index 04721cfa4..d0bbfe33c 100644
--- a/tokio-threadpool/Cargo.toml
+++ b/tokio-threadpool/Cargo.toml
@@ -22,10 +22,9 @@ categories = ["concurrency", "asynchronous"]
 [dependencies]
 tokio-executor = { version = "0.1.2", path = "../tokio-executor" }
 futures = "0.1.19"
-crossbeam = "0.6.0"
-crossbeam-channel = "0.3.3"
-crossbeam-deque = "0.6.1"
-crossbeam-utils = "0.6.2"
+crossbeam-deque = "0.7.0"
+crossbeam-queue = "0.1.0"
+crossbeam-utils = "0.6.4"
 num_cpus = "1.2"
 rand = "0.6"
 slab = "0.4.1"
diff --git a/tokio-threadpool/src/builder.rs b/tokio-threadpool/src/builder.rs
index 9c49135e0..82c7cac65 100644
--- a/tokio-threadpool/src/builder.rs
+++ b/tokio-threadpool/src/builder.rs
@@ -3,7 +3,6 @@ use config::{Config, MAX_WORKERS};
 use park::{BoxPark, BoxedPark, DefaultPark};
 use shutdown::ShutdownTrigger;
 use pool::{Pool, MAX_BACKUP};
-use task::Queue;
 use thread_pool::ThreadPool;
 use worker::{self, Worker, WorkerId};
 
@@ -13,6 +12,7 @@ use std::sync::Arc;
 use std::time::Duration;
 use std::cmp::max;
 
+use crossbeam_deque::Injector;
 use num_cpus;
 use tokio_executor::Enter;
 use tokio_executor::park::Park;
@@ -414,7 +414,7 @@ impl Builder {
             workers.into()
         };
 
-        let queue = Arc::new(Queue::new());
+        let queue = Arc::new(Injector::new());
 
         // Create a trigger that will clean up resources on shutdown.
         //
diff --git a/tokio-threadpool/src/lib.rs b/tokio-threadpool/src/lib.rs
index 3340cca9f..45ee6626a 100644
--- a/tokio-threadpool/src/lib.rs
+++ b/tokio-threadpool/src/lib.rs
@@ -79,9 +79,8 @@
 
 extern crate tokio_executor;
 
-extern crate crossbeam;
-extern crate crossbeam_channel;
-extern crate crossbeam_deque as deque;
+extern crate crossbeam_deque;
+extern crate crossbeam_queue;
 extern crate crossbeam_utils;
 #[macro_use]
 extern crate futures;
diff --git a/tokio-threadpool/src/pool/mod.rs b/tokio-threadpool/src/pool/mod.rs
index 3d7a93773..2326fca04 100644
--- a/tokio-threadpool/src/pool/mod.rs
+++ b/tokio-threadpool/src/pool/mod.rs
@@ -15,7 +15,7 @@ use self::backup_stack::BackupStack;
 
 use config::Config;
 use shutdown::ShutdownTrigger;
-use task::{Blocking, Queue, Task};
+use task::{Blocking, Task};
 use worker::{self, Worker, WorkerId};
 
 use futures::Poll;
@@ -27,6 +27,7 @@ use std::sync::atomic::AtomicUsize;
 use std::sync::{Arc, Weak};
 use std::thread;
 
+use crossbeam_deque::Injector;
 use crossbeam_utils::CachePadded;
 use rand;
 
@@ -57,7 +58,7 @@ pub(crate) struct Pool {
     //
     // Spawned tasks are pushed into this queue. Although worker threads have their own dedicated
     // task queues, they periodically steal tasks from this global queue, too.
-    pub queue: Arc<Queue>,
+    pub queue: Arc<Injector<Arc<Task>>>,
 
     // Completes the shutdown process when the `ThreadPool` and all `Worker`s get dropped.
     //
@@ -90,7 +91,7 @@ impl Pool {
         trigger: Weak<ShutdownTrigger>,
         max_blocking: usize,
         config: Config,
-        queue: Arc<Queue>,
+        queue: Arc<Injector<Arc<Task>>>,
     ) -> Pool {
         let pool_size = workers.len();
         let total_size = max_blocking + pool_size;
diff --git a/tokio-threadpool/src/shutdown.rs b/tokio-threadpool/src/shutdown.rs
index 880fc8220..290cb182c 100644
--- a/tokio-threadpool/src/shutdown.rs
+++ b/tokio-threadpool/src/shutdown.rs
@@ -1,6 +1,7 @@
-use task::Queue;
+use task::Task;
 use worker;
 
+use crossbeam_deque::Injector;
 use futures::{Future, Poll, Async};
 use futures::task::AtomicTask;
 
@@ -62,14 +63,17 @@ impl Future for Shutdown {
 pub(crate) struct ShutdownTrigger {
     inner: Arc<Mutex<Inner>>,
     workers: Arc<[worker::Entry]>,
-    queue: Arc<Queue>,
+    queue: Arc<Injector<Arc<Task>>>,
 }
 
 unsafe impl Send for ShutdownTrigger {}
 unsafe impl Sync for ShutdownTrigger {}
 
 impl ShutdownTrigger {
-    pub(crate) fn new(workers: Arc<[worker::Entry]>, queue: Arc<Queue>) -> ShutdownTrigger {
+    pub(crate) fn new(
+        workers: Arc<[worker::Entry]>,
+        queue: Arc<Injector<Arc<Task>>>,
+    ) -> ShutdownTrigger {
         ShutdownTrigger {
             inner: Arc::new(Mutex::new(Inner {
                 task: AtomicTask::new(),
@@ -84,7 +88,7 @@ impl ShutdownTrigger {
 impl Drop for ShutdownTrigger {
     fn drop(&mut self) {
         // Drain the global task queue.
-        while self.queue.pop().is_some() {}
+        while !self.queue.steal().is_empty() {}
 
         // Drop the remaining incomplete tasks and parkers assosicated with workers.
         for worker in self.workers.iter() {
diff --git a/tokio-threadpool/src/task/mod.rs b/tokio-threadpool/src/task/mod.rs
index 90592f203..fe535b30d 100644
--- a/tokio-threadpool/src/task/mod.rs
+++ b/tokio-threadpool/src/task/mod.rs
@@ -1,10 +1,8 @@
 mod blocking;
 mod blocking_state;
-mod queue;
 mod state;
 
 pub(crate) use self::blocking::{Blocking, CanBlock};
-pub(crate) use self::queue::Queue;
 use self::blocking_state::BlockingState;
 use self::state::State;
 
diff --git a/tokio-threadpool/src/task/queue.rs b/tokio-threadpool/src/task/queue.rs
deleted file mode 100644
index 166438e78..000000000
--- a/tokio-threadpool/src/task/queue.rs
+++ /dev/null
@@ -1,34 +0,0 @@
-use task::Task;
-
-use std::sync::Arc;
-
-use crossbeam_channel::{unbounded, Receiver, Sender};
-
-#[derive(Debug)]
-pub(crate) struct Queue {
-    // TODO(stjepang): Use a custom, faster MPMC queue implementation that supports `steal_many()`.
-    chan: (Sender<Arc<Task>>, Receiver<Arc<Task>>),
-}
-
-// ===== impl Queue =====
-
-impl Queue {
-    /// Create a new, empty, `Queue`.
-    pub fn new() -> Queue {
-        Queue {
-            chan: unbounded(),
-        }
-    }
-
-    /// Push a task onto the queue.
-    #[inline]
-    pub fn push(&self, task: Arc<Task>) {
-        self.chan.0.send(task).unwrap();
-    }
-
-    /// Pop a task from the queue.
-    #[inline]
-    pub fn pop(&self) -> Option<Arc<Task>> {
-        self.chan.1.try_recv().ok()
-    }
-}
diff --git a/tokio-threadpool/src/worker/entry.rs b/tokio-threadpool/src/worker/entry.rs
index 07bee5bd2..e3a363276 100644
--- a/tokio-threadpool/src/worker/entry.rs
+++ b/tokio-threadpool/src/worker/entry.rs
@@ -9,9 +9,9 @@ use std::sync::atomic::{AtomicBool, AtomicUsize, Ordering};
 use std::sync::atomic::Ordering::{Acquire, AcqRel, Relaxed, Release};
 use std::time::Duration;
 
-use crossbeam::queue::SegQueue;
+use crossbeam_deque::{Steal, Stealer, Worker};
+use crossbeam_queue::SegQueue;
 use crossbeam_utils::CachePadded;
-use deque;
 use slab::Slab;
 
 // TODO: None of the fields should be public
@@ -29,10 +29,10 @@ pub(crate) struct WorkerEntry {
     next_sleeper: UnsafeCell<usize>,
 
     // Worker half of deque
-    worker: deque::Worker<Arc<Task>>,
+    pub worker: Worker<Arc<Task>>,
 
     // Stealer half of deque
-    stealer: deque::Stealer<Arc<Task>>,
+    stealer: Stealer<Arc<Task>>,
 
     // Thread parker
     park: UnsafeCell<Option<BoxPark>>,
@@ -53,7 +53,8 @@ pub(crate) struct WorkerEntry {
 
 impl WorkerEntry {
     pub fn new(park: BoxPark, unpark: BoxUnpark) -> Self {
-        let (w, s) = deque::fifo();
+        let w = Worker::new_fifo();
+        let s = w.stealer();
 
         WorkerEntry {
             state: CachePadded::new(AtomicUsize::new(State::default().into())),
@@ -187,7 +188,7 @@ impl WorkerEntry {
     /// This **must** only be called by the thread that owns the worker entry.
     /// This function is not `Sync`.
     #[inline]
-    pub fn pop_task(&self) -> deque::Pop<Arc<Task>> {
+    pub fn pop_task(&self) -> Option<Arc<Task>> {
         self.worker.pop()
     }
 
@@ -199,23 +200,15 @@ impl WorkerEntry {
     /// At the same time, this method steals some additional tasks and moves
     /// them into `dest` in order to balance the work distribution among
     /// workers.
-    pub fn steal_tasks(&self, dest: &Self) -> deque::Steal<Arc<Task>> {
-        self.stealer.steal_many(&dest.worker)
+    pub fn steal_tasks(&self, dest: &Self) -> Steal<Arc<Task>> {
+        self.stealer.steal_batch_and_pop(&dest.worker)
     }
 
     /// Drain (and drop) all tasks that are queued for work.
     ///
     /// This is called when the pool is shutting down.
     pub fn drain_tasks(&self) {
-        use deque::Pop::*;
-
-        loop {
-            match self.worker.pop() {
-                Data(_) => {}
-                Empty => break,
-                Retry => {}
-            }
-        }
+        while self.worker.pop().is_some() {}
     }
 
     /// Parks the worker thread.
@@ -284,7 +277,6 @@ impl WorkerEntry {
         }
         running_tasks.clear();
 
-        // Drop the parker.
         unsafe {
             *self.park.get() = None;
             *self.unpark.get() = None;
@@ -297,7 +289,7 @@ impl WorkerEntry {
         if self.needs_drain.compare_and_swap(true, false, Acquire) {
             let running_tasks = unsafe { &mut *self.running_tasks.get() };
 
-            while let Some(task) = self.remotely_completed_tasks.try_pop() {
+            while let Ok(task) = self.remotely_completed_tasks.pop() {
                 running_tasks.remove(task.reg_index.get());
             }
         }
diff --git a/tokio-threadpool/src/worker/mod.rs b/tokio-threadpool/src/worker/mod.rs
index e270c10d9..939d0063f 100644
--- a/tokio-threadpool/src/worker/mod.rs
+++ b/tokio-threadpool/src/worker/mod.rs
@@ -386,16 +386,13 @@ impl Worker {
     ///
     /// Returns `true` if work was found.
     fn try_run_owned_task(&self, notify: &Arc<Notifier>) -> bool {
-        use deque::Pop;
-
         // Poll the internal queue for a task to run
         match self.entry().pop_task() {
-            Pop::Data(task) => {
+            Some(task) => {
                 self.run_task(task, notify);
                 true
             }
-            Pop::Empty => false,
-            Pop::Retry => true,
+            None => false,
         }
     }
 
@@ -403,7 +400,7 @@ impl Worker {
     ///
     /// Returns `true` if work was found
     fn try_steal_task(&self, notify: &Arc<Notifier>) -> bool {
-        use deque::Steal;
+        use crossbeam_deque::Steal;
 
         debug_assert!(!self.is_blocking.get());
 
@@ -415,7 +412,7 @@ impl Worker {
         loop {
             if idx < len {
                 match self.pool.workers[idx].steal_tasks(self.entry()) {
-                    Steal::Data(task) => {
+                    Steal::Success(task) => {
                         trace!("stole task from another worker");
 
                         self.run_task(task, notify);
@@ -701,15 +698,17 @@ impl Worker {
     ///
     /// Returns `true` if this worker has tasks in its queue.
     fn sleep_light(&self) {
-        const STEAL_COUNT: usize = 32;
-
         self.entry().park_timeout(Duration::from_millis(0));
 
-        for _ in 0..STEAL_COUNT {
-            if let Some(task) = self.pool.queue.pop() {
-                self.pool.submit(task, &self.pool);
-            } else {
-                break;
+        use crossbeam_deque::Steal;
+        loop {
+            match self.pool.queue.steal_batch(&self.entry().worker) {
+                Steal::Success(()) => {
+                    self.pool.signal_work(&self.pool);
+                    break;
+                }
+                Steal::Empty => break,
+                Steal::Retry => {}
             }
         }
     }