Because of the influence of the space local scattering, when two non-coherent desired signals are incident at the same time, the beampattern of the traditional adaptive beamformer can form the null in one of the desired signal directions, which destroys the integrity of the mainlobe and leads the output signal-to-interference-plus-noise ratio to drop significantly. In response to this, a robust adaptive beamforming method based on the mainlobe shape-preserving is proposed. The method processes the covariance matrix by the diagonal loading technique. And the optimal loading factor is determined by constructing the model of maximizing the output signal-to-interference-plus-noise ratio of the one of the two desired signals under the condition that the output signal-to-interference-plus-noise ratio of another desired signal is not less than a certain value. The optimal loading factor obtained by the method is not only simple to calculate, but also has a closed solution, which is easy to be applied in the practical application. Simulation results show that the method not only avoids forming the nulls in both of the desired signal directions and keeps the shape of the mainlobe, but also improves the array output signal-to-interference-plus-noise ratio, so as to achieve the purpose of the robust adaptive beamforming.