In this project, we investigate computer vision techniques to analyze expression in 3D facial image both with and without pose variations. 3D image not only contains more information than conventional 2D data, it is also less affected by illumination and head pose variation. However, capture, construction, and classification of 3D facial images suffer from several real-world challenges. Due to head pose, facial occlusion, improper stereo images, the outcome of 3D reconstruction of facial images can be erroneous subjected to loss of data. These affect the classification accuracy in 3D facial images. We are investigating different novel geometric invariant features and machine learning techniques that are robust to such real world constraints in order to recognize different facial expressions. Our current focus is invariant recognition of these expressions under 3D pose variations.