A network constructed by arcs and vertices is a useful tool to model a real-life system, such as a computer/communication, an electric power transmission, a transportation, or a logistics system. Network reliability, the probability to satisfy customers' demand, is a common performance index of such a system. From the quality of service viewpoint, the network reliability optimization is an important objective which many enterprises pursue to maintain their customer satisfaction. Combining with the characteristic of assignment problem, this study is mainly to search for the optimal components assignment with maximal network reliability. A set of components is ready to be assigned to the arcs, each component may have several modes, such as failure, maintenance, or partially reserved by other customers, and thus the network according to any component assignment is multistate. A minimal-cut based genetic algorithm is developed to solve such a problem. In order to show the computational efficiency of the proposed algorithm, a simple computer network and a real one are adopted while comparing with the implicit enumeration method and the approach of random solution generation, respectively.