diff --git a/deps/amqp_client/test/network_client_SUITE.erl b/deps/amqp_client/test/network_client_SUITE.erl
index 04b35b378e..2f04b9207f 100644
--- a/deps/amqp_client/test/network_client_SUITE.erl
+++ b/deps/amqp_client/test/network_client_SUITE.erl
@@ -65,6 +65,12 @@ teardown_test() ->
 rpc_test() ->
     test_util:rpc_test(new_connection()).
 
+pub_and_close_test() ->
+    {timeout, 60,
+        fun() ->
+            test_util:pub_and_close_test(new_connection(), new_connection())
+        end}.
+
 %%---------------------------------------------------------------------------
 %% Negative Tests
 
diff --git a/deps/amqp_client/test/test_util.erl b/deps/amqp_client/test/test_util.erl
index 91e1e3bbdf..d115382dfb 100644
--- a/deps/amqp_client/test/test_util.erl
+++ b/deps/amqp_client/test/test_util.erl
@@ -307,15 +307,6 @@ basic_reject_test(Connection) ->
 %% So, in the end, if we set up a consumer, we should get all the important
 %% messages, but not all spam (we waited for all messages in buffer to be sent,
 %% but didn't allow further adding to the buffer).
-%% This test sometimes fails due to other causes - if a message leaves the
-%% socket doesn't mean it will be delivered; or maybe all spam messages have
-%% actually been buffered before the connection closed, thus all of them will
-%% be delivered.
-%% ... and sometimes succeedes (when it shouldn't) - the messages might be
-%% buffered and delivered very quickly (but not the spam), before the
-%% connection gets to close.
-%% ... though in the end, per a larger number of tests, the greater of the
-%% success and failure rates reflects the real results.
 %% First connection is used for sending messages, second connection is used to
 %% get the messages.
 pub_and_close_test(Connection1, Connection2) ->