diff --git a/Lib/test/regrtest.py b/Lib/test/regrtest.py
--- a/Lib/test/regrtest.py
+++ b/Lib/test/regrtest.py
@@ -940,37 +940,34 @@ def runtest(test, verbose, quiet,
     use_timeout = (timeout is not None)
     if use_timeout:
         faulthandler.dump_traceback_later(timeout, exit=True)
     try:
         support.match_tests = match_tests
         if failfast:
             support.failfast = True
         if output_on_failure:
-            support.verbose = True
-
             # Reuse the same instance to all calls to runtest(). Some
             # tests keep a reference to sys.stdout or sys.stderr
             # (eg. test_argparse).
             if runtest.stringio is None:
                 stream = io.StringIO()
                 runtest.stringio = stream
             else:
                 stream = runtest.stringio
                 stream.seek(0)
                 stream.truncate()
 
             orig_stdout = sys.stdout
             orig_stderr = sys.stderr
             try:
                 sys.stdout = stream
                 sys.stderr = stream
-                result = runtest_inner(test, verbose, quiet, huntrleaks,
-                                       display_failure=False)
-                if result[0] == FAILED:
+                result = runtest_inner(test, verbose, quiet, huntrleaks)
+                if verbose or result[0] == FAILED:
                     output = stream.getvalue()
                     orig_stderr.write(output)
                     orig_stderr.flush()
             finally:
                 sys.stdout = orig_stdout
                 sys.stderr = orig_stderr
         else:
             support.verbose = verbose  # Tell tests to be moderately quiet