In this paper, we use the fuzzy stochastic technique to predict the mainframe's CPU utilization from the given database of IBM's OS/390 system log at a specific time. Consequently, system programmers can tune the system to improve its performance by moving out the unimportant jobs during the peak time. To address this issue, we proposed an effective method, a fuzzy stochastic prediction (FSP), to predict the CPU utilization in a specific time. The fuzzy stochastic variable is a parameter in FSP, which is generated by the recent system behavior.