To solve the contradiction between the information completeness and data volume existing in the radar signal fusion，a signal-level fusion algorithm for radar emitter identification based on sparse representation is presented. In the algorithm, intercepted signals are projected to sparse domains and compressed for communication. Then the sparse coefficients are fused for recognition. The data of communication is decreased in the algorithm, and the available information of intercepted signal is also reserved. The simulations show that the recognition accuracy is improved compared with the single signal receiver and the decision-level fusion algorithms.