In linear representation based face recognition (FR), it is expected that a discriminative dictionary can be learned from the training samples so that the query sample can be better represented for classification. On the other hand, dimensionality reduction is also an important issue for FR. It cannot only reduce significantly the storage space of face images, but also enhance the discrimination of face feature. Existing methods mostly perform dimensionality reduction and dictionary learning separately, which may not fully exploit the discriminative information in the training samples. In this paper, we propose to learn jointly the projection matrix for dimensionality reduction and the discriminative dictionary for face representation. The joint learning makes the learned projection and dictionary better fit with each other so that a more effective face classification can be obtained. The proposed algorithm is evaluated on benchmark face databases in comparison with existing linear representation based methods, and the results show that the joint learning improves the FR rate, particularly when the number of training samples per class is small. ; Department of Computing