The multi-compartments vehicle routing problem is not only an NP-hard problem in strong sense, but also has multiple constraints. In addition to the classical capacity and time-window constraints, this problem also has compartment, compatibility, and assignment constraints. For such a problem, a solution found by traditional heuristics usually can be improved by using a searching scheme. In this paper, a hybrid genetic algorithm for the multi-compartments vehicle routing problem, that uses the cluster-first route-second heuristic proposed by Lin, Tseng, Chen, and Tsai (2017) to generate the initial population and to design the related genetic operators, is proposed.