diff --git a/Doc/library/multiprocessing.rst b/Doc/library/multiprocessing.rst
index 2b67d10d7bf1b7..23aee321a6a7e6 100644
--- a/Doc/library/multiprocessing.rst
+++ b/Doc/library/multiprocessing.rst
@@ -869,10 +869,13 @@ Note that one can also create a shared queue by using a manager object -- see
    bother you then you can instead use a queue created with a
    :ref:`manager <multiprocessing-managers>`.
 
-   (1) After putting an object on an empty queue there may be an
-       infinitesimal delay before the queue's :meth:`~Queue.empty`
-       method returns :const:`False` and :meth:`~Queue.get_nowait` can
-       return without raising :exc:`queue.Empty`.
+   (1) After putting an object on an empty queue there may be a delay
+       before :meth:`~Queue.get_nowait` can return without raising
+       :exc:`queue.Empty`, because the feeder thread flushes objects to
+       the underlying pipe asynchronously. On platforms where
+       ``sem_getvalue()`` is not implemented (for example macOS), the
+       queue's :meth:`~Queue.empty` method may also remain :const:`True`
+       during this delay.
 
    (2) If multiple processes are enqueuing objects, it is possible for
        the objects to be received at the other end out-of-order.
@@ -947,8 +950,17 @@ For an example of the usage of queues for interprocess communication see
       Return ``True`` if the queue is empty, ``False`` otherwise.  Because of
       multithreading/multiprocessing semantics, this is not reliable.
 
+      On platforms where ``sem_getvalue()`` is implemented, this method
+      uses the same approximate size accounting as :meth:`~Queue.qsize`.
+      Otherwise, it may report ``True`` while items are still buffered and
+      waiting to be flushed to the underlying pipe.
+
       May raise an :exc:`OSError` on closed queues. (not guaranteed)
 
+      .. versionchanged:: 3.15
+         On platforms where ``sem_getvalue()`` is implemented, this method
+         now uses semaphore-based queue size accounting.
+
    .. method:: full()
 
       Return ``True`` if the queue is full, ``False`` otherwise.  Because of
diff --git a/Lib/multiprocessing/queues.py b/Lib/multiprocessing/queues.py
index 981599acf5ef26..d7bfc6ca0665f0 100644
--- a/Lib/multiprocessing/queues.py
+++ b/Lib/multiprocessing/queues.py
@@ -124,7 +124,17 @@ def qsize(self):
         return self._maxsize - self._sem.get_value()
 
     def empty(self):
-        return not self._poll()
+        # Preserve the historical "closed queue may raise OSError" behavior.
+        # q.close() is a no-op for unused queues, so this only raises once the
+        # reader end has actually been closed.
+        if self._closed:
+            self._poll()
+
+        try:
+            return self._sem.get_value() == self._maxsize
+        except NotImplementedError:
+            # Fallback for platforms without sem_getvalue() (for example macOS).
+            return not self._poll()
 
     def full(self):
         return self._sem._semlock._is_zero()
diff --git a/Lib/test/_test_multiprocessing.py b/Lib/test/_test_multiprocessing.py
index cc07062eee6f98..9ba81be2c1b5fa 100644
--- a/Lib/test/_test_multiprocessing.py
+++ b/Lib/test/_test_multiprocessing.py
@@ -1235,7 +1235,18 @@ def test_get(self):
                 break
         self.assertEqual(queue_empty(queue), False)
 
-        self.assertEqual(queue.get_nowait(), 1)
+        for _ in support.sleeping_retry(support.SHORT_TIMEOUT):
+            try:
+                value = queue.get_nowait()
+            except pyqueue.Empty:
+                # Queue.empty() may become false before the feeder thread
+                # flushes objects to the pipe.
+                continue
+            else:
+                break
+        else:
+            self.fail("queue.get_nowait() unexpectedly raised Empty")
+        self.assertEqual(value, 1)
         self.assertEqual(queue.get(True, None), 2)
         self.assertEqual(queue.get(True), 3)
         self.assertEqual(queue.get(timeout=1), 4)
@@ -1320,6 +1331,29 @@ def test_qsize(self):
         self.assertEqual(q.qsize(), 0)
         close_queue(q)
 
+    def test_empty_uses_semaphore_count(self):
+        if self.TYPE != 'processes':
+            self.skipTest(f'test not appropriate for {self.TYPE}')
+
+        q = self.Queue()
+        try:
+            q._sem.get_value()
+        except NotImplementedError:
+            close_queue(q)
+            self.skipTest('sem_getvalue not implemented on this platform')
+
+        q.put('sentinel')
+        original_poll = q._poll
+        q._poll = lambda timeout=0.0: False
+        try:
+            self.assertFalse(q.empty())
+        finally:
+            q._poll = original_poll
+
+        self.assertEqual(q.get(timeout=support.SHORT_TIMEOUT), 'sentinel')
+        self.assertTrue(q.empty())
+        close_queue(q)
+
     @classmethod
     def _test_task_done(cls, q):
         for obj in iter(q.get, None):
diff --git a/Misc/NEWS.d/next/Library/2026-02-16-11-20-00.gh-issue-142837.xP4kLm.rst b/Misc/NEWS.d/next/Library/2026-02-16-11-20-00.gh-issue-142837.xP4kLm.rst
new file mode 100644
index 00000000000000..69bc73c5e5c282
--- /dev/null
+++ b/Misc/NEWS.d/next/Library/2026-02-16-11-20-00.gh-issue-142837.xP4kLm.rst
@@ -0,0 +1,4 @@
+Fixed :meth:`multiprocessing.Queue.empty` to use semaphore-based size
+accounting on platforms that support ``sem_getvalue()``, making its behavior
+consistent with :meth:`multiprocessing.Queue.qsize` and
+:meth:`multiprocessing.Queue.full`.