This paper presents a novel descriptor named Region based Extensive Response Index Pattern (RETRaIN) for facial expression recognition. The RETRaIN encodes the relation among the reference and neighboring pixels of facial active regions. These relations are computed by using directional compass mask on an input image and extract the high edge responses in foremost directions. Further extreme edge index positions are selected and encoded into six-bit compact code to reduce feature dimensionality and distinguish between the uniform and non-uniform patterns in the facial features. The performance of the proposed descriptor is tested and evaluated on three benchmark datasets Extended Cohn Kanade, JAFFE, and MUG. The RETRaIN achieves superior recognition accuracy in comparison to state-of-the-art techniques.