Resource matching, a procedure of Grid task scheduling, aims at finding appropriate resources for Grid tasks according to its requirements. The resource information recorded in Grid Resource Information Service （GRIS） is always static. Because of dynamic characteristic of local jobs arrived in a Grid site, that some of the resources in candidate resource set determined based on static information can not satisfy tasks QoS requirements. This paper predicted resource future status based on its historical dynamic information. Grid task average completion time and distribution of completion time were given. According to tasks QoS requirements and considering both current and future status of a resource, a resource matching model and a matching algorithm were presented. The experimental results show that the algorithm can reduce amount of candidate resource effectively, which can decrease time complexity of a scheduling.