diff --git a/Output-format.md b/Output-format.md
new file mode 100644
index 0000000..14d5415
--- /dev/null
+++ b/Output-format.md
@@ -0,0 +1,22 @@
+You can optionally output the estimated information to a .csv file. We intended to have similar output data with [Pupil Labs](https://docs.pupil-labs.com/).
+
+## CSV file
+The content of output .csv file are following:
+* `timestamp` - timestamp of the source image frame
+* `index` - frame number of input video or camera
+* `face_id` - face identification generated by OpenFace with tracking
+* `confidence` - face and facial landmark detection certainty ranges from `-1` to `1`, provided by OpenFace. A value of `-1` indicates no confidence. `1` indicates perfect confidence.
+* `gaze_2d_x` - x position in the estimated gaze in normalized coordinates. The coordinate is based on the screen such that the origin `(0,0)` is the **top left** and `(1,1)` is the **bottom right**
+* `gaze_2d_y` - y position in the estimated gaze in normalized coordinates. The coordinate system is the same as for `norm_pose_y`
+* `face_center_3d_x` - x center of face in the camera coordinate system
+* `face_center_3d_y` - y center of face in the camera coordinate system
+* `face_center_3d_z` - z center of face in the camera coordinate system
+* `gaze_3d_x` - x position of the 3D gaze direction vector in the camera coordinate system
+* `gaze_3d_y` - y position of the 3D gaze direction vector in the camera coordinate system
+* `gaze_3d_z` - z position of the 3D gaze direction vector in the camera coordinate system
+* `left_eye_3d_x` - x center of left eye in the camera coordinate system
+* `left_eye_3d_y` - y center of left eye in the camera coordinate system
+* `left_eye_3d_z` - z center of left eye in the camera coordinate system
+* `right_eye_3d_x` - x center of right eye in the camera coordinate system
+* `right_eye_3d_y` - y center of right eye in the camera coordinate system
+* `right_eye_3d_z` - z center of right eye in the camera coordinate system