The assembly accuracy of multistage rotors is crucial to guarantee the quality and performance of aero engines. However, traditional assembly accuracy analysis methods usually oversimplify geometric deviations as the degrees of freedom (DOF) variations relative to the ideal position, ignoring real surface morphology and the resulting non-uniform contact deformation. Therefore, this paper proposes an assembly accuracy analysis method that integrates the effects of surface morphology and non-uniform contact deformation on error propagation process, and further adopts a phase optimization method to improve the assembly accuracy of multistage rotors. Specifically, the surface morphology of rotor end faces is generated using the random midpoint displacement method. The Conjugate Gradient-Fast Fourier Transform (CG-FFT) algorithm is employed to calculate the non-uniform contact deformation. According to the simulation results, a fitting deformation surface (FDS) is constructed to determine the assembly position and orientation. On this basis, a novel error propagation model is established and the mounting phase of each stage rotor is optimized by a genetic algorithm (GA). A case study of a specific rotor system is carried out to verify the proposed error propagation model and give the optimized assembly combination with the objective of minimum coaxiality. This research effectively improves the prediction and actual assembly accuracy of aero-engine multistage rotors and provide a theoretical basis for the precision assembly of multistage rotors in aero-engine applications.