Research on automatic techniques for analyzing human facial behavior has been intensified, driven by the emerging research field of affective computing (R. Picard, 2003) and its important application in human computer interaction and intelligent systems. The face surface is a three dimensional time-varying `wave', which is associated with the movement of facial expressions. Tracing the behavior of the 3D primitive features could reveal precious information about the nature of the underlying physical process. In this paper, we propose to study the intensity-variant facial expressions in a 3D space. We present a new approach for 3D expression model analysis, tracking and classification. We demonstrate the feasibility and advantage of the proposed method using the 3D range data for prototypic 3D facial expression recognition