多处理器系统互连网络的子网络可靠性是衡量系统性能的一个关键指标。为了刻画（n，k）-星图中（n-m，k-m）-星图子网络的容错性能，在概率故障条件下分析了（n，k）-星图中存在无故障（n-m，k-m）-星图子网络的概率。对于1≤k≤n -1和1≤m≤k -1，得出了（n-m，k-m）-星图子网络存在概率的上下界的理论公式，给出了仅考虑点故障的（n，k）-星图的无故障（n-m，k-m）-星图子网络的搜寻算法，并基于蒙特卡罗模拟提出了评估无故障（n-m，k-m）-星图子网络的存在概率的近似方法。实验结果表明，随着顶点可靠性逐渐变小，得出的（n-m，k-m）-星图子网络存在概率的上下界与近似评估结果基本吻合；当顶点可靠性较高或（n-m，k-m）-星图子网络存在概率的上下界相差较大时，利用基于蒙特卡罗的近似方法可以得出较为精确的评估结果。

The subnetwork reliability of the interconnection network of a multiprocessor system is a key indicator to assess the performance of the multiprocessor system. In order to characterize the fault tolerance of (n-m,k-m)-star graph subnetworks in an (n,k)-star graph, the probability of existing fault-free (n-m,k-m)-star graph subnetworks in an (n,k)-star graph under probabilistic fault condition is analysed. For 1≤k≤n -1 and 1≤m≤k -1, the theoretical formulas of the upper and lower bounds of the probability of existing (n-m,k-m)-star graph subnetworks are obtained, and an algorithm for searching fault-free (n-m,k-m)-star graph subnetworks in an (n,k)-star graph with only node failures is given. Moreover, an approximate method for evaluating the existence probability of fault-free (n-m,k-m)-star graph subnetworks is given based on Monte Carlo simulation. The experimental results show that the upper and lower bounds of the probability of existing (n-m,k-m)-star graph subnetworks are basically consistent with the approximate evaluation results as the node reliability gradually becomes smaller, and the relatively accurate evaluation result can be obtained by the approximate method based on Monte Carlo simulation when the node reliability is relatively high or the difference between the upper and lower bounds of the probability of existing (n-m,k-m)-star graph subnetworks is significant.