diff --git a/CHANGELOG.rst b/CHANGELOG.rst
index 15a028dd..ac21f5d4 100644
--- a/CHANGELOG.rst
+++ b/CHANGELOG.rst
@@ -2,6 +2,17 @@
 Changelog
 =========
 
+* Fixed the terminal summary contradicting the exit code when ``--cov-fail-under`` is used.
+
+  The exit code is decided with coverage.py's ``should_fail_under``, which compares the total rounded to
+  ``[report] precision``, but the summary banner compared the raw total. When the raw total was below the
+  threshold and rounded up to meet it, the run exited 0 while printing a red
+  ``FAIL Required test coverage of N% not reached``.
+  The banner now uses the same predicate as the exit code, so ``[report] precision`` is honoured in both.
+  See `#638 <https://github.com/pytest-dev/pytest-cov/issues/638>`_
+  (`#728 <https://github.com/pytest-dev/pytest-cov/issues/728>`_ reports the same inconsistency
+  from the opposite direction).
+
 7.1.0 (2026-03-21)
 ------------------
 
diff --git a/src/pytest_cov/plugin.py b/src/pytest_cov/plugin.py
index 9f5c3f76..d70294b9 100644
--- a/src/pytest_cov/plugin.py
+++ b/src/pytest_cov/plugin.py
@@ -409,8 +409,14 @@ def pytest_terminal_summary(self, terminalreporter):
             terminalreporter.write(report)
 
         if self.options.cov_fail_under is not None and self.options.cov_fail_under > 0:
+            # import coverage lazily here to avoid importing
+            # it for unit tests that don't need it
+            from coverage.results import should_fail_under
+
             self.write_heading(terminalreporter)
-            failed = self.cov_total < self.options.cov_fail_under
+            # Use the same predicate that decided the exit code in
+            # pytest_sessionfinish, so this banner cannot contradict it.
+            failed = should_fail_under(self.cov_total, self.options.cov_fail_under, self.options.cov_precision)
             markup = {'red': True, 'bold': True} if failed else {'green': True}
             message = '{fail}Required test coverage of {required}% {reached}. Total coverage: {actual:.2f}%\n'.format(
                 required=self.options.cov_fail_under,
diff --git a/tests/test_pytest_cov.py b/tests/test_pytest_cov.py
index b291f432..1a0f40a7 100644
--- a/tests/test_pytest_cov.py
+++ b/tests/test_pytest_cov.py
@@ -535,6 +535,31 @@ def test_cov_min_float_value_not_reached(testdir):
     result.stdout.fnmatch_lines(['FAIL Required test coverage of 88.89% not reached. Total coverage: 88.89%'])
 
 
+def test_cov_min_float_value_rounds_up_to_threshold(testdir):
+    # The summary banner must agree with the exit code. SCRIPT covers
+    # 88.888...%, which rounds to 89% at the default precision of 0, so the
+    # run passes. The banner used to compare the raw total instead and
+    # printed a red FAIL on a passing run. See #638 and #728.
+    script = testdir.makepyfile(SCRIPT)
+
+    result = testdir.runpytest('-v', f'--cov={script.dirpath()}', '--cov-report=term-missing', '--cov-fail-under=88.9', script)
+    assert result.ret == 0
+    result.stdout.fnmatch_lines(['Required test coverage of 88.9% reached. Total coverage: 88.89%'])
+
+
+def test_cov_min_float_value_rounds_up_to_threshold_with_precision(testdir):
+    # Same divergence, driven by an explicit [report] precision as in #638.
+    script = testdir.makepyfile(SCRIPT)
+    testdir.tmpdir.join('.coveragerc').write("""
+[report]
+precision = 1
+""")
+
+    result = testdir.runpytest('-v', f'--cov={script.dirpath()}', '--cov-report=term-missing', '--cov-fail-under=88.9', script)
+    assert result.ret == 0
+    result.stdout.fnmatch_lines(['Required test coverage of 88.9% reached. Total coverage: 88.89%'])
+
+
 def test_cov_min_float_value_not_reached_cli(testdir):
     script = testdir.makepyfile(SCRIPT)
     result = testdir.runpytest(