Three-dimensional solid element models often with a great number of degrees-of-freedom (DOFs) are now widely used for rotor dynamic analysis. While without reduction, it will cost considerable calculating resources and time to solve the equations of motion, especially when Monte Carlo simulation (MCS) is needed for stochastic analysis. To improve the analysis efficiency, the DOFs are partly reduced to modal spaces, and the stochastic results (critical speeds or unbalance response) are expanded to polynomial spaces. First, a reduced rotor model is got by component mode synthesis (CMS), and the stochastic results are expanded by polynomial chaos basis with unknown coefficients. Then, the reduced rotor model is used to calculate the sample results to obtain the coefficients. At last, the expressions of the result by polynomial chaos basis are used as surrogate models for MCS. An aero-engine rotor system with uncertain parameters is analyzed. The accuracy of the method is validated by direct MCS, and the high efficiency makes it possible for stochastic dynamic analysis of complex engine rotor systems modeled by 3D solid element.