Abstract. The paper develops a model to evaluate the availability and the cost optimization for a multi-states series-parallel system, whose components are subject to periodic preventive maintenance. The objective is to optimize for each system component the maintenance policy minimizing a cost function of the system, under the constraint of required availability and for a specified period. This policy identifies the good times of components to perform preventive maintenance, as well as the dates of the first inspections. We consider that system cost is a cost of preventive maintenance. The universal generating function model is used to assess the performance distribution of the entire system and the system availability. The optimization is done for different values of required availability. The effect of required availability on the preventive maintenance policy and system cost is studied. The optimization technique is based on the genetic algorithm. An illustrative example is presented.