In this paper we proposed a joint spatial and temporal ICA method for face recognition, and compared the performances of different ICA approaches (spatiotemporal ICA and spatial ICA). In our study, two face datasets collected by AcSys FRS Discovery system were used. One face dataset involves less variation in terms of face expression and head movement, while the other encompasses much more change. The experimental results led to the following conclusions: 1) The number of features affects the recognition rate; 2) Spatiotemporal ICA outperforms spatial ICA in every scenarios; 3) The type of classifier is also a fact that affects the recognition rate. These findings justify the promise of spatiotemporal ICA for face recognition.