diff --git a/moto/autoscaling/models.py b/moto/autoscaling/models.py
index 24811be73..83de9712a 100644
--- a/moto/autoscaling/models.py
+++ b/moto/autoscaling/models.py
@@ -684,6 +684,18 @@ class AutoScalingBackend(BaseBackend):
         for instance in protected_instances:
             instance.protected_from_scale_in = protected_from_scale_in
 
+    def notify_terminate_instances(self, instance_ids):
+        for autoscaling_group_name, autoscaling_group in self.autoscaling_groups.items():
+            original_instance_count = len(autoscaling_group.instance_states)
+            autoscaling_group.instance_states = list(filter(
+                lambda i_state: i_state.instance.id not in instance_ids,
+                autoscaling_group.instance_states
+            ))
+            difference = original_instance_count - len(autoscaling_group.instance_states)
+            if difference > 0:
+                autoscaling_group.replace_autoscaling_group_instances(difference, autoscaling_group.get_propagated_tags())
+                self.update_attached_elbs(autoscaling_group_name)
+
 
 autoscaling_backends = {}
 for region, ec2_backend in ec2_backends.items():
diff --git a/moto/ec2/responses/instances.py b/moto/ec2/responses/instances.py
index 3f73d2e94..82c2b1997 100644
--- a/moto/ec2/responses/instances.py
+++ b/moto/ec2/responses/instances.py
@@ -1,5 +1,7 @@
 from __future__ import unicode_literals
 from boto.ec2.instancetype import InstanceType
+
+from moto.autoscaling import autoscaling_backends
 from moto.core.responses import BaseResponse
 from moto.core.utils import camelcase_to_underscores
 from moto.ec2.utils import filters_from_querystring, \
@@ -65,6 +67,7 @@ class InstanceResponse(BaseResponse):
         instance_ids = self._get_multi_param('InstanceId')
         if self.is_not_dryrun('TerminateInstance'):
             instances = self.ec2_backend.terminate_instances(instance_ids)
+            autoscaling_backends[self.region].notify_terminate_instances(instance_ids)
             template = self.response_template(EC2_TERMINATE_INSTANCES)
             return template.render(instances=instances)
 
diff --git a/tests/test_autoscaling/test_autoscaling.py b/tests/test_autoscaling/test_autoscaling.py
index 750605c07..a0e5c395f 100644
--- a/tests/test_autoscaling/test_autoscaling.py
+++ b/tests/test_autoscaling/test_autoscaling.py
@@ -1269,3 +1269,36 @@ def test_set_desired_capacity_down_boto3():
     instance_ids = {instance['InstanceId'] for instance in group['Instances']}
     set(protected).should.equal(instance_ids)
     set(unprotected).should_not.be.within(instance_ids)  # only unprotected killed
+
+
+@mock_autoscaling
+@mock_ec2
+def test_terminate_instance_in_autoscaling_group():
+    mocked_networking = setup_networking()
+    client = boto3.client('autoscaling', region_name='us-east-1')
+    _ = client.create_launch_configuration(
+        LaunchConfigurationName='test_launch_configuration'
+    )
+    _ = client.create_auto_scaling_group(
+        AutoScalingGroupName='test_asg',
+        LaunchConfigurationName='test_launch_configuration',
+        MinSize=1,
+        MaxSize=20,
+        VPCZoneIdentifier=mocked_networking['subnet1'],
+        NewInstancesProtectedFromScaleIn=False
+    )
+
+    response = client.describe_auto_scaling_groups(AutoScalingGroupNames=['test_asg'])
+    original_instance_id = next(
+        instance['InstanceId']
+        for instance in response['AutoScalingGroups'][0]['Instances']
+    )
+    ec2_client = boto3.client('ec2', region_name='us-east-1')
+    ec2_client.terminate_instances(InstanceIds=[original_instance_id])
+
+    response = client.describe_auto_scaling_groups(AutoScalingGroupNames=['test_asg'])
+    replaced_instance_id = next(
+        instance['InstanceId']
+        for instance in response['AutoScalingGroups'][0]['Instances']
+    )
+    replaced_instance_id.should_not.equal(original_instance_id)