A new scheduling problem in which a job requiring more than one machine at the same time has appeared as multiprocessor systems have developed. In this paper, the fixed-multiprocessor-task-scheduling problem including the precedence constraint is studied. The purpose is to schedule jobs to minimize makespan. Kramer solves this problem optimally by branch and bound algorithm. Since his method is too time intensive, we solve this problem by genetic algorithm (GA). Since the only way to reduce makespan in this problem is to process parallel-relationship jobs at the same moment of time, a mutation based on parallel relationship is proposed. In the numerical example, the results of genetic algorithm are compared with those of Kramer`s branch and bound algorithm.