In this paper, the assembly error transfer prediction model was established based on the error transfer attribute and the error transfer path of the assembly process. The assembly accuracy prediction was realized by combining error variation inequality, constraint inequality and mathematical statistics method. Aiming at the lowest processing cost of high-precision equipment assembly, the assembly accuracy required the reliability of index, the value interval of tolerance item and the principle of tolerance selection as constraints, the tolerance optimization model was established. The tolerance optimization of high-precision equipment assembly was carried out by combining the penalty function method and the improved genetic algorithm. Finally, taking gear oil pump assembly as an example, according to the reliability calculation formula of assembly accuracy, the reliability is 97.35%. Compared with the processing cost before and after optimization, the processing cost decreased by 12.13% after optimization, and the assembly accuracy reliability increased by 1.65%. The correctness and effectiveness of assembly accuracy prediction and tolerance optimization are verified.