A novel Active Discriminant Functions (ADFs) for handwriting recognition is presented in this paper. First, statistical feature based deformable model in principal subspace is proposed and a minimum distance between an unknown pattern and the deformable model is given. Second, to improve the accuracy of recognition, the minor subspace is also considered in ADFs. Third, as parameters of the ADFs, the optimal constraints of a deformable model are searched by applying Minimum Classification Error (MCE) criterion. Finally, empirical experiments are conducted on handwritten Chinese characters used in banking and the results show that our proposed ADFs outperform other representative techniques, such as support vector machine, multiplayer perceptron, etc.