Sparse representation for classification (SRC) has attracted much attention in the field of pattern recognition in recent years. If each class has few training samples, SRC usually cannot achieve the desirable recognition performance. To address the above problem, extended sparse representation for classification (ESRC) is proposed,which uses the intraclass variant matrix to supplement the training sample information. Nevertheless, ESRC can hardly capture the nonlinear information in complex data such as images. An extended sparse representation in a feature space for classification algorithm was proposed, in which the original data were mapped into a new high dimensional space through a nonlinear mapping to evaluate the contribution of each training sample in the representation of test sample, and each sample was given a certain weight according to the contribution. Then, the proposed algorithm used the weighted training samples combining the intraclass variant matrix to represent the test samples. Experiments show that the proposed algorithm is superior to other typical sparse representation for classification algorithms.