Mixed-model assembly lines are become more and more important by producing different models of the same product on an assembly line. Aiming at the existing mixed-model assembly line balancing problem, first, two important objective functions for minimizing cycle time and workload variance were provided, and mathematical models were established. Furthermore, in order to obtain the optimal or near optimal solutions, an improved genetic algorithm was proposed with combined precedence graph. Finally, the experiment results illustrate the feasibility and validity of the proposed improved genetic algorithm.