Abstract: Preemption of tasks guarantees the schedulability of system but, on the other hand, the overheads relevant to preemption ate unavoidable. For the embedded real-time systems of which the resource is constrained, it is important to control the frequency of task preemption so as to reduce the resource consumption due to preemption. A preemption control model based on time slice is therefore proposed to assign every task on upper limit of preemptable frequency. To minimize the upper limit under the constraint on schedulability, the genetic algorithm is used to optimize it. Then, the algorithm for optimization is analyzed by simulation, and the result shows that the preemption control strategy proposed can reduce the frequency of preemption effectively during task execution.