A result verification procedure is required to guarantee the correctness of task results executed by any unspecified resource in desktop grids. Voting-based and trust-based schemes have typically been used for desktop grid. However, these schemes can encounter two problems: waste of resources, due to redundant replications of each task, and increased turnaround time, due to the inability to deal with a dynamic changeable execution environment. To overcome these problems, we propose a fuzzy inference-based replication scheme that can adaptively determine the number of replications per task needed for result verification. Simulation results indicate that our scheme is superior to others in terms of turnaround time of entire tasks, quantity of resources consumed, and number of reallocations per task.