A computer-implemented technique is described for verifying the identity of a user using two components of face analysis. In a first part, the technique determines whether captured face information matches a previously stored structural face signature pertaining to the user. The structural face signature describes, at least in part, gross structural characteristics of the face that are largely invariant from pose to pose. In the second part, the technique determines whether the captured face information matches a dynamic face signature associated with the user. The dynamic face signature describes movement of parts of the face over a span of time as the user performs a gesture, and the correlation of different parts of the face during the movement. The technique reduces the risk that a malicious actor can successfully artificially duplicate the appearance of an authorized user.