diff --git a/test/e2e_node/eviction_test.go b/test/e2e_node/eviction_test.go
index a190f737e72..b78a5411ea8 100644
--- a/test/e2e_node/eviction_test.go
+++ b/test/e2e_node/eviction_test.go
@@ -51,7 +51,7 @@ import (
 const (
 	postTestConditionMonitoringPeriod = 1 * time.Minute
 	evictionPollInterval              = 2 * time.Second
-	pressureDisappearTimeout          = 1 * time.Minute
+	pressureDisappearTimeout          = 10 * time.Minute
 	// pressure conditions often surface after evictions because the kubelet only updates
 	// node conditions periodically.
 	// we wait this period after evictions to make sure that we wait out this delay
diff --git a/test/e2e_node/system_node_critical_test.go b/test/e2e_node/system_node_critical_test.go
index 0c5b21e5d74..fcecccaf28f 100644
--- a/test/e2e_node/system_node_critical_test.go
+++ b/test/e2e_node/system_node_critical_test.go
@@ -38,6 +38,14 @@ var _ = SIGDescribe("SystemNodeCriticalPod [Slow] [Serial] [Disruptive] [NodeFea
 	// this test only manipulates pods in kube-system
 	f.SkipNamespaceCreation = true
 
+	ginkgo.AfterEach(func() {
+		if framework.TestContext.PrepullImages {
+			// The test may cause the prepulled images to be evicted,
+			// prepull those images again to ensure this test not affect following tests.
+			PrePullAllImages()
+		}
+	})
+
 	ginkgo.Context("when create a system-node-critical pod", func() {
 		tempSetCurrentKubeletConfig(f, func(initialConfig *kubeletconfig.KubeletConfiguration) {
 			diskConsumed := resource.MustParse("200Mi")