Recently,the phenomenon of software aging,one in which error conditions actually accrue with time and/or load,has been observed. To counteract software aging, which can cause outages resulting in high costs,a proactive restart technique called software rejuvenation is proposed and the rejuvenation cost is analyzed.In order to reduce the rejuvenation cost and improve software availability and reliability further, rejuvenation granularity should be finer than before. Therefore a fine-grained proactive technique——multilevel software rejuvenation is put forward. Firstly, the degradation law of system performance can be determined by analyzing the occupation and wastage of system resources. Based on the law and the software architecture,the two software rejuvenation policies, i.e. time-based multilevel software rejuvenation policy and detection-based multilevel software rejuvenation policy, can be drafted, and the rejuvenation granularity can be determined. Their formal description of policies is given by finite-state automaton. Finally, the entire process is illustrated with a web service case. This paper provides a case to illustrate the process, and the simulation results of the case show that the multilevel software rejuvenation policy can reduce the MTTR and rejuvenation cost further, comparing with the only system-level software rejuvenation. As a consequence,the system availability and reliability are enhanced.