diff --git a/ceilometer/tests/pipeline_base.py b/ceilometer/tests/pipeline_base.py
index 826a1169f..846594ca7 100644
--- a/ceilometer/tests/pipeline_base.py
+++ b/ceilometer/tests/pipeline_base.py
@@ -1219,6 +1219,42 @@ class BasePipelineTestCase(test.BaseTestCase):
         self.assertEqual(expected_length, len(publisher.samples))
         return sorted(publisher.samples, key=lambda s: s.volume)
 
+    def test_aggregator_meter_type(self):
+        volumes = [1.0, 2.0, 3.0]
+        transformer_cfg = [
+            {
+                'name': 'aggregator',
+                'parameters': {'size': len(volumes) * len(sample.TYPES)}
+            },
+        ]
+        self._set_pipeline_cfg('transformers', transformer_cfg)
+        self._set_pipeline_cfg('counters',
+                               ['testgauge', 'testcumulative', 'testdelta'])
+        counters = []
+        for sample_type in sample.TYPES:
+            for volume in volumes:
+                counters.append(sample.Sample(
+                    name='test' + sample_type,
+                    type=sample_type,
+                    volume=volume,
+                    unit='B',
+                    user_id='test_user',
+                    project_id='test_proj',
+                    resource_id='test_resource',
+                    timestamp=timeutils.utcnow().isoformat(),
+                    resource_metadata={'version': '1.0'}
+                ))
+
+        pipeline_manager = pipeline.PipelineManager(self.pipeline_cfg,
+                                                    self.transformer_manager)
+        pipe = pipeline_manager.pipelines[0]
+
+        pipe.publish_samples(None, counters)
+        pipe.flush(None)
+        publisher = pipeline_manager.pipelines[0].publishers[0]
+        actual = sorted(s.volume for s in publisher.samples)
+        self.assertEqual([2.0, 3.0, 6.0], actual)
+
     def test_aggregator_input_validation(self):
         aggregator = conversions.AggregatorTransformer("1", "15", None,
                                                        None, None)
diff --git a/ceilometer/transformer/conversions.py b/ceilometer/transformer/conversions.py
index 5e1592b7a..956a5586f 100644
--- a/ceilometer/transformer/conversions.py
+++ b/ceilometer/transformer/conversions.py
@@ -15,6 +15,7 @@
 # License for the specific language governing permissions and limitations
 # under the License.
 
+import collections
 import re
 
 import six
@@ -165,6 +166,7 @@ class AggregatorTransformer(ScalingTransformer):
                  **kwargs):
         super(AggregatorTransformer, self).__init__(**kwargs)
         self.samples = {}
+        self.counts = collections.defaultdict(int)
         self.size = int(size) if size else None
         self.retention_time = float(retention_time) if retention_time else None
         self.initial_timestamp = None
@@ -200,23 +202,27 @@ class AggregatorTransformer(ScalingTransformer):
         # NOTE(sileht): it assumes, a meter always have the same unit/type
         return "%s-%s-%s" % (s.name, s.resource_id, non_aggregated_keys)
 
-    def handle_sample(self, context, sample):
+    def handle_sample(self, context, sample_):
         if not self.initial_timestamp:
-            self.initial_timestamp = timeutils.parse_isotime(sample.timestamp)
+            self.initial_timestamp = timeutils.parse_isotime(sample_.timestamp)
 
         self.aggregated_samples += 1
-        key = self._get_unique_key(sample)
+        key = self._get_unique_key(sample_)
+        self.counts[key] += 1
         if key not in self.samples:
-            self.samples[key] = self._convert(sample)
+            self.samples[key] = self._convert(sample_)
             if self.merged_attribute_policy[
                     'resource_metadata'] == 'drop':
                 self.samples[key].resource_metadata = {}
         else:
-            self.samples[key].volume += self._scale(sample)
+            if sample_.type == sample.TYPE_CUMULATIVE:
+                self.samples[key].volume = self._scale(sample_)
+            else:
+                self.samples[key].volume += self._scale(sample_)
             for field in self.merged_attribute_policy:
                 if self.merged_attribute_policy[field] == 'last':
                     setattr(self.samples[key], field,
-                            getattr(sample, field))
+                            getattr(sample_, field))
 
     def flush(self, context):
         if not self.initial_timestamp:
@@ -228,7 +234,13 @@ class AggregatorTransformer(ScalingTransformer):
         full = self.aggregated_samples >= self.size
         if full or expired:
             x = self.samples.values()
-            self.samples = {}
+            # gauge aggregates need to be averages
+            for s in x:
+                if s.type == sample.TYPE_GAUGE:
+                    key = self._get_unique_key(s)
+                    s.volume /= self.counts[key]
+            self.samples.clear()
+            self.counts.clear()
             self.aggregated_samples = 0
             self.initial_timestamp = None
             return x