In this paper,  a generativescorespace model was proposed, based on which a face recognition approach was derived. First, the proposed approach designed a probabilistic generative model for face representation, which effectively combined the flexibility of partsbased paradigm   with the robustness of sparse component analysis. Then, a score function (i.e. feature mapping) was derived based on the model. Besides,  a similarity measure was constructed for single sample face identification, which is essentially the function over observed variables, hidden variables and model parameters. The proposed approach was evaluated on two standard face databases to validate its effectiveness.