In recent years,a number of direction-finding algorithms for noncircular signals are proposed.The asymptotic performance and Cramer-Rao Bound (CRB) have been analyzed in the literatures.However,the robustness of these algorithms with modeling errors is out of range of the published papers.In this paper,the MUSIC algorithm for noncircular signals is introduced.The closed-form expression of mean square error is obtained from the first-order perturbation expansion of the spatial spectrum function.The simulation results validate our analysis.The robustness to the angular separation and the noncircularity phase with modeling errors is analyzed by virtue of the analytical results.