This paper proposes an approach to evaluating the product assembly precision in different assembly sequences, considering the effect of joint surface deformation. Three assembly variation sources, including manufacturing variation, assembly clearance and joint surface deformation, which have effect on the final assembly precision are analyzed. Based on finite element analysis, a hybrid genetic algorithm and back-propagation neural network model is built to predict the assembly variations, which are caused by the joint surface deformation under different assembly conditions and different parameters of the joint surface. An assembly variation propagation model is built, and a product assembly precision evaluation approach is proposed to identify the feasible assembly sequences, and the optimal assembly sequence considering the effect of the joint surface deformation. Finally, a case study is given to verify the proposed approach.