diff --git a/ml_engine/online_prediction/README.md b/ml_engine/online_prediction/README.md
new file mode 100644
index 00000000000..07c160f2311
--- /dev/null
+++ b/ml_engine/online_prediction/README.md
@@ -0,0 +1 @@
+https://cloud.google.com/ml-engine/docs/concepts/prediction-overview
diff --git a/ml_engine/online_prediction/predict.py b/ml_engine/online_prediction/predict.py
new file mode 100644
index 00000000000..5727177420a
--- /dev/null
+++ b/ml_engine/online_prediction/predict.py
@@ -0,0 +1,198 @@
+#!/bin/python
+# Copyright 2017 Google Inc. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Examples of using the Cloud ML Engine's online prediction service."""
+import argparse
+import base64
+import json
+
+# [START import_libraries]
+import googleapiclient.discovery
+# [END import_libraries]
+import six
+
+
+# [START predict_json]
+def predict_json(project, model, instances, version=None):
+    """Send json data to a deployed model for prediction.
+
+    Args:
+        project (str): project where the Cloud ML Engine Model is deployed.
+        model (str): model name.
+        instances ([Mapping[str: Any]]): Keys should be the names of Tensors
+            your deployed model expects as inputs. Values should be datatypes
+            convertible to Tensors, or (potentially nested) lists of datatypes
+            convertible to tensors.
+        version: str, version of the model to target.
+    Returns:
+        Mapping[str: any]: dictionary of prediction results defined by the
+            model.
+    """
+    # Create the ML Engine service object.
+    # To authenticate set the environment variable
+    # GOOGLE_APPLICATION_CREDENTIALS=<path_to_service_account_file>
+    service = googleapiclient.discovery.build('ml', 'v1beta1')
+    name = 'projects/{}/models/{}'.format(project, model)
+
+    if version is not None:
+        name += '/versions/{}'.format(version)
+
+    response = service.projects().predict(
+        name=name,
+        body={'instances': instances}
+    ).execute()
+
+    if 'error' in response:
+        raise RuntimeError(response['error'])
+
+    return response['predictions']
+# [END predict_json]
+
+
+# [START predict_tf_records]
+def predict_tf_records(project,
+                       model,
+                       example_bytes_list,
+                       version=None):
+    """Send protocol buffer data to a deployed model for prediction.
+
+    Args:
+        project (str): project where the Cloud ML Engine Model is deployed.
+        model (str): model name.
+        example_bytes_list ([str]): A list of bytestrings representing
+            serialized tf.train.Example protocol buffers. The contents of this
+            protocol buffer will change depending on the signature of your
+            deployed model.
+        version: str, version of the model to target.
+    Returns:
+        Mapping[str: any]: dictionary of prediction results defined by the
+            model.
+    """
+    service = googleapiclient.discovery.build('ml', 'v1beta1')
+    name = 'projects/{}/models/{}'.format(project, model)
+
+    if version is not None:
+        name += '/versions/{}'.format(version)
+
+    response = service.projects().predict(
+        name=name,
+        body={'instances': [
+            {'b64': base64.b64encode(example_bytes)}
+            for example_bytes in example_bytes_list
+        ]}
+    ).execute()
+
+    if 'error' in response:
+        raise RuntimeError(response['error'])
+
+    return response['predictions']
+# [END predict_tf_records]
+
+
+# [START census_to_example_bytes]
+def census_to_example_bytes(json_instance):
+    """Serialize a JSON example to the bytes of a tf.train.Example.
+    This method is specific to the signature of the Census example.
+    See: https://cloud.google.com/ml-engine/docs/concepts/prediction-overview
+    for details.
+
+    Args:
+        json_instance (Mapping[str: Any]): Keys should be the names of Tensors
+            your deployed model expects to parse using it's tf.FeatureSpec.
+            Values should be datatypes convertible to Tensors, or (potentially
+            nested) lists of datatypes convertible to tensors.
+    Returns:
+        str: A string as a container for the serialized bytes of
+            tf.train.Example protocol buffer.
+    """
+    import tensorflow as tf
+    feature_dict = {}
+    for key, data in json_instance.iteritems():
+        if isinstance(data, six.string_types):
+            feature_dict[key] = tf.train.Feature(
+                bytes_list=tf.train.BytesList(value=[str(data)]))
+        elif isinstance(data, float):
+            feature_dict[key] = tf.train.Feature(
+                float_list=tf.train.FloatList(value=[data]))
+        elif isinstance(data, int):
+            feature_dict[key] = tf.train.Feature(
+                int64_list=tf.train.Int64List(value=[data]))
+    return tf.train.Example(
+        features=tf.train.Features(
+            feature=feature_dict
+        )
+    ).SerializeToString()
+# [END census_to_example_bytes]
+
+
+def main(project, model, version=None, force_tfrecord=False):
+    """Send user input to the prediction service."""
+    while True:
+        try:
+            user_input = json.loads(raw_input("Valid JSON >>>"))
+        except KeyboardInterrupt:
+            return
+
+        if not isinstance(user_input, list):
+            user_input = [user_input]
+        try:
+            if force_tfrecord:
+                example_bytes_list = [
+                    census_to_example_bytes(e)
+                    for e in user_input
+                ]
+                result = predict_tf_records(
+                    project, model, example_bytes_list, version=version)
+            else:
+                result = predict_json(
+                    project, model, user_input, version=version)
+        except RuntimeError as err:
+            print(str(err))
+        else:
+            print(result)
+
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        '--project',
+        help='Project in which the model is deployed',
+        type=str,
+        required=True
+    )
+    parser.add_argument(
+        '--model',
+        help='Model name',
+        type=str,
+        required=True
+    )
+    parser.add_argument(
+        '--version',
+        help='Name of the version.',
+        type=str
+    )
+    parser.add_argument(
+        '--force-tfrecord',
+        help='Send predictions as TFRecords rather than raw JSON',
+        action='store_true',
+        default=False
+    )
+    args = parser.parse_args()
+    main(
+        args.project,
+        args.model,
+        version=args.version,
+        force_tfrecord=args.force_tfrecord
+    )
diff --git a/ml_engine/online_prediction/predict_test.py b/ml_engine/online_prediction/predict_test.py
new file mode 100644
index 00000000000..90930cae611
--- /dev/null
+++ b/ml_engine/online_prediction/predict_test.py
@@ -0,0 +1,74 @@
+# Copyright 2017 Google Inc. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Tests for predict.py ."""
+
+import base64
+
+import pytest
+
+import predict
+
+
+MODEL = 'census'
+VERSION = 'v1'
+TF_RECORDS_VERSION = 'v1tfrecord'
+PROJECT = 'python-docs-samples-tests'
+JSON = {
+    'age': 25,
+    'workclass': ' Private',
+    'education': ' 11th',
+    'education_num': 7,
+    'marital_status': ' Never-married',
+    'occupation': ' Machine-op-inspct',
+    'relationship': ' Own-child',
+    'race': ' Black',
+    'gender': ' Male',
+    'capital_gain': 0,
+    'capital_loss': 0,
+    'hours_per_week': 40,
+    'native_country': ' United-States'
+}
+EXPECTED_OUTPUT = {
+    u'probabilities': [0.9942260384559631, 0.005774002522230148],
+    u'logits': [-5.148599147796631],
+    u'classes': 0,
+    u'logistic': [0.005774001590907574]
+}
+
+
+def test_predict_json():
+    result = predict.predict_json(
+        PROJECT, MODEL, [JSON, JSON], version=VERSION)
+    assert [EXPECTED_OUTPUT, EXPECTED_OUTPUT] == result
+
+
+def test_predict_json_error():
+    with pytest.raises(RuntimeError):
+        predict.predict_json(PROJECT, MODEL, [{"foo": "bar"}], version=VERSION)
+
+
+@pytest.mark.slow
+def test_census_example_to_bytes():
+    b = predict.census_to_example_bytes(JSON)
+    assert base64.b64encode(b) is not None
+
+
+@pytest.mark.slow
+@pytest.mark.xfail('Single placeholder inputs broken in service b/35778449')
+def test_predict_tfrecords():
+    b = predict.census_to_example_bytes(JSON)
+    result = predict.predict_tfrecords(
+        PROJECT, MODEL, [b, b], version=TF_RECORDS_VERSION)
+    assert [EXPECTED_OUTPUT, EXPECTED_OUTPUT] == result
diff --git a/ml_engine/online_prediction/requirements.txt b/ml_engine/online_prediction/requirements.txt
new file mode 100644
index 00000000000..1ecc5960e54
--- /dev/null
+++ b/ml_engine/online_prediction/requirements.txt
@@ -0,0 +1 @@
+tensorflow==1.0.0