diff --git a/pkg/controller/podautoscaler/replica_calculator.go b/pkg/controller/podautoscaler/replica_calculator.go
index 2ee7fefdb68..07067ff98eb 100644
--- a/pkg/controller/podautoscaler/replica_calculator.go
+++ b/pkg/controller/podautoscaler/replica_calculator.go
@@ -237,9 +237,15 @@ func (c *ReplicaCalculator) calcPlainMetricReplicas(metrics metricsclient.PodMet
 		return currentReplicas, utilization, nil
 	}
 
+	newReplicas := int32(math.Ceil(newUsageRatio * float64(len(metrics))))
+	if (newUsageRatio < 1.0 && newReplicas > currentReplicas) || (newUsageRatio > 1.0 && newReplicas < currentReplicas) {
+		// return the current replicas if the change of metrics length would cause a change in scale direction
+		return currentReplicas, utilization, nil
+	}
+
 	// return the result, where the number of replicas considered is
 	// however many replicas factored into our calculation
-	return int32(math.Ceil(newUsageRatio * float64(len(metrics)))), utilization, nil
+	return newReplicas, utilization, nil
 }
 
 // GetObjectMetricReplicas calculates the desired replica count based on a target metric utilization (as a milli-value)
diff --git a/pkg/controller/podautoscaler/replica_calculator_test.go b/pkg/controller/podautoscaler/replica_calculator_test.go
index dc683334000..88b18549741 100644
--- a/pkg/controller/podautoscaler/replica_calculator_test.go
+++ b/pkg/controller/podautoscaler/replica_calculator_test.go
@@ -1481,6 +1481,22 @@ func TestReplicaCalcDuringRollingUpdateWithMaxSurge(t *testing.T) {
 	tc.runTest(t)
 }
 
+func TestReplicaCalcDuringRollingUpdateWithMaxSurgeCM(t *testing.T) {
+	tc := replicaCalcTestCase{
+		currentReplicas:  2,
+		expectedReplicas: 2,
+		podPhase:         []v1.PodPhase{v1.PodRunning, v1.PodRunning, v1.PodRunning},
+		metric: &metricInfo{
+			name:                "qps",
+			levels:              []int64{10000, 10000},
+			targetUtilization:   17000,
+			expectedUtilization: 10000,
+			metricType:          podMetric,
+		},
+	}
+	tc.runTest(t)
+}
+
 // TestComputedToleranceAlgImplementation is a regression test which
 // back-calculates a minimal percentage for downscaling based on a small percentage
 // increase in pod utilization which is calibrated against the tolerance value.