diff --git a/templates/zephyr/index.html b/templates/zephyr/index.html
index ac65b1f752..3c0d909cd9 100644
--- a/templates/zephyr/index.html
+++ b/templates/zephyr/index.html
@@ -50,6 +50,7 @@
 
 {# Not escaped, because it's guaranteed by the model to be an integer. #}
 var initial_pointer = {{ user_profile.pointer }};
+var server_generation = {{ server_generation }};
 
 var email = "{{ user_profile.user.email|escapejs }}";
 var have_initial_messages = {{ have_initial_messages|escapejs }};
diff --git a/tools/jslint/check-all.js b/tools/jslint/check-all.js
index 4c8113fbf4..04e6c75da8 100644
--- a/tools/jslint/check-all.js
+++ b/tools/jslint/check-all.js
@@ -7,6 +7,7 @@ var globals =
 
     // index.html
     + ' initial_pointer email stream_list people_list have_initial_messages'
+    + ' server_generation'
 
     // compose.js
     + ' show_compose hide_compose toggle_compose clear_compose_box compose_button'
diff --git a/zephyr/static/js/zephyr.js b/zephyr/static/js/zephyr.js
index eea4dfd053..b062495e19 100644
--- a/zephyr/static/js/zephyr.js
+++ b/zephyr/static/js/zephyr.js
@@ -57,9 +57,12 @@ var selected_message;  // = get_message_row(selected_message_id)
 var received = {
     first: -1,
     last:  -1,
-    failures: 0
+    failures: 0,
+    server_generation: -1 /* to be filled in on document.ready */
 };
 
+$(function () { received.server_generation = server_generation; });
+
 // The "message groups", i.e. blocks of messages collapsed by recipient.
 // Each message table has a list of lists.
 var message_groups = {
diff --git a/zephyr/views.py b/zephyr/views.py
index 85084394f3..a416a9afbf 100644
--- a/zephyr/views.py
+++ b/zephyr/views.py
@@ -28,6 +28,9 @@ import socket
 import re
 import hashlib
 import urllib
+import time
+
+SERVER_GENERATION = int(time.time())
 
 def require_post(view_func):
     def _wrapped_view_func(request, *args, **kwargs):
@@ -174,7 +177,8 @@ def home(request):
                                'have_initial_messages':
                                    'true' if num_messages > 0 else 'false',
                                'show_debug':
-                                   settings.DEBUG and ('show_debug' in request.GET) },
+                                   settings.DEBUG and ('show_debug' in request.GET),
+                               'server_generation': SERVER_GENERATION},
                               context_instance=RequestContext(request))
 
 @login_required
@@ -203,12 +207,14 @@ def format_updates_response(messages, mit_sync_bot=False, apply_markdown=False,
     return {'messages': [message.to_dict(apply_markdown) for message in messages],
             "result": "success",
             "msg": "",
-            'where':   where}
+            'where':   where,
+            'server_generation': SERVER_GENERATION}
 
 def return_messages_immediately(request, handler, user_profile, **kwargs):
     first = request.POST.get("first")
     last = request.POST.get("last")
     failures = request.POST.get("failures")
+    client_server_generation = request.POST.get("server_generation")
     if first is None or last is None:
         # When an API user is first querying the server to subscribe,
         # there's no reason to reply immediately.
@@ -251,6 +257,12 @@ def return_messages_immediately(request, handler, user_profile, **kwargs):
         handler.finish(format_updates_response([], where="bottom", **kwargs))
         return True
 
+    if client_server_generation is not None and int(client_server_generation) != SERVER_GENERATION:
+        # No messages, but still return immediately to inform the
+        # client that they should reload
+        handler.finish(format_updates_response([], where="bottom", **kwargs))
+        return True
+
     return False
 
 def get_updates_backend(request, user_profile, handler, **kwargs):