Abstract: Considering that prior knowledge has errors in practical application, a worst-case performance optimization beamforming algorithm with invariable weight vector length is presented instead of the conventional one that is linearly constrained. Analyzes the mathematical similarity between the neural minor component analysis (MCA) learning rule and beamforming optimization problem. Then, the neural MCA learning rule is used to implement robust adaptive beamforming. Computer simulations show that the proposed algorithm has stronger signal trackability and higher resistance to interference, and it is robuster in the presence of signal steering vector errors in comparison with the linearly constrained beamforming algorithm.