In practical applications, multiple types of array errors exist simultaneously. In view of the orientation dependence of array errors in this case, an improved MUSIC(multiple signal classification) algorithm based on MST(manifold separation technique) was proposed, which can effectively solve DOA(direction of arrival) estimation problem under the influence of multiple array errors. The sampling matrix, which contains all the non-ideal characteristics of the array was obtained by using MST, so as to achieve accurate direction finding. By using two-dimensional Fourier transform, the 2D spatial spectrum was solved. Compared with existing MUSIC algorithm, the computation amount of spectrum peak searching was reduced. Theoretical analysis and simulation verify the effectiveness of the algorithm, which can provide reference for the solution of practical problems.