diff --git a/builtin/providers/aws/resource_aws_autoscaling_group.go b/builtin/providers/aws/resource_aws_autoscaling_group.go
index 17110e70c67c..bbefbc36ef42 100644
--- a/builtin/providers/aws/resource_aws_autoscaling_group.go
+++ b/builtin/providers/aws/resource_aws_autoscaling_group.go
@@ -154,6 +154,11 @@ func resourceAwsAutoscalingGroup() *schema.Resource {
 				Set:      schema.HashString,
 			},
 
+			"delete_delay": &schema.Schema{
+				Type:     schema.TypeInt,
+				Optional: true,
+			},
+
 			"metrics_granularity": &schema.Schema{
 				Type:     schema.TypeString,
 				Optional: true,
@@ -624,6 +629,20 @@ func getAwsAutoscalingGroup(
 	return nil, nil
 }
 
+func resourceAwsAutoscalingScaleDownToCount(d *schema.ResourceData, count int64, conn *autoscaling.AutoScaling) error {
+	log.Printf("[DEBUG] Reducing autoscaling group capacity to %d", count)
+	opts := autoscaling.UpdateAutoScalingGroupInput{
+		AutoScalingGroupName: aws.String(d.Id()),
+		MinSize:              aws.Int64(0),
+		DesiredCapacity:      aws.Int64(count),
+		MaxSize:              aws.Int64(count),
+	}
+	if _, err := conn.UpdateAutoScalingGroup(&opts); err != nil {
+		return fmt.Errorf("Error setting capacity to %d to drain: %s", count, err)
+	}
+	return nil
+}
+
 func resourceAwsAutoscalingGroupDrain(d *schema.ResourceData, meta interface{}) error {
 	conn := meta.(*AWSClient).autoscalingconn
 
@@ -632,20 +651,36 @@ func resourceAwsAutoscalingGroupDrain(d *schema.ResourceData, meta interface{})
 		return nil
 	}
 
-	// First, set the capacity to zero so the group will drain
-	log.Printf("[DEBUG] Reducing autoscaling group capacity to zero")
-	opts := autoscaling.UpdateAutoScalingGroupInput{
-		AutoScalingGroupName: aws.String(d.Id()),
-		DesiredCapacity:      aws.Int64(0),
-		MinSize:              aws.Int64(0),
-		MaxSize:              aws.Int64(0),
-	}
-	if _, err := conn.UpdateAutoScalingGroup(&opts); err != nil {
-		return fmt.Errorf("Error setting capacity to zero to drain: %s", err)
+	if delay, ok := d.GetOk("delete_delay"); ok {
+		g, err := getAwsAutoscalingGroup(d.Id(), conn)
+		if err != nil {
+			return err
+		}
+		delayT := time.Duration(delay.(int)) * time.Second
+		t := time.Now()
+		count := *g.DesiredCapacity
+		for count > 0 {
+			count--
+			if err := resourceAwsAutoscalingScaleDownToCount(d, count, conn); err != nil {
+				return err
+			}
+			if count != 0 {
+				now := time.Now()
+				sleepTime := delayT - now.Sub(t)
+				t = now
+				if sleepTime > 0 {
+					time.Sleep(sleepTime)
+				}
+			}
+		}
+	} else {
+		if err := resourceAwsAutoscalingScaleDownToCount(d, 0, conn); err != nil {
+			return err
+		}
 	}
 
 	// Next, wait for the autoscale group to drain
-	log.Printf("[DEBUG] Waiting for group to have zero instances")
+	log.Printf("[DEBUG] Waiting for group to have 0 instances")
 	return resource.Retry(10*time.Minute, func() *resource.RetryError {
 		g, err := getAwsAutoscalingGroup(d.Id(), conn)
 		if err != nil {