The MVDR beamformer has better spatial resolution and better interference rejection capability than the standard beamformer, provided that the signal steering vector is accurately known. However, in practical applications, the performance of MVDR beamformer degrades severely because of violation of underlying assumptions on the environment, sensor array and sample covariance matrix errors. To overcome the mismatches, a robust MVDR beamforming algorithm based on the diagonal loading was proposed. By considering the bad effect of steering vector errors on MVDR algorithm performance, and obtaining the optimal weight vector under the condition of the largest deviation, this algorithm can sufficiently reduce the effect of the mismatches and provide an improved robustness in the communication environment. Also, the algorithm uses a recursion process to avoid inversing matrix, which reduces the computational complexity and facilitates project implementation. The simulation results show that the proposed robust algorithm has better performance than the traditional MVDR algorithm.