The knapsack problem (KP) is a classical NP problem. The multi-objective knapsack problem (MKP) is more difficult than KP. The fitness vector function is firstly introduced, that can solve the problem of non-convex solutions, which is difficult for the common aggregation function method. Then, with the master-slave multi-population cooperation method, the cooperation between the global exploration and local development is improved, so the multi-objective multi-population cooperative GA (MMGA) posed in this paper enhances also the optimization ability of GA. The experiment result shows, MMGA is superior to NSGAII for MKP. Furthermore, with the both of master-slave multi-population cooperation and the mechanism of holding non-dominate solutions, the premature and degradation phenomena are effectively prevent.