It has been previously demonstrated that systems based on Hidden Markov Models (HMMs) are suitable for face recognition. The proposed approaches in the literature are either HMMs with one-dimensional (1D-HMMs) or two-dimensional (2D-HMMs) topology. Both have shown some serious drawbacks. The 1D-HMM approaches typically use a whole row (or column) of an image as observation vector and by consequence do not allow horizontal (or vertical) alignment. 2D-HMM approaches present some implementation issues because of the computational cost. In this paper, we propose a 1D-HMM approach which allow the use of local features and we will demonstate the accuracy of this approach on the so-called BANCA database.