Web caching is a scalable and effective way to reduce network traffic and response time. In this study, we propose a new cache replacement policy for proxy and Web servers, the Size-Adjusted Sliding Window LFU (SSW-LFU). In this policy, we use the rates of recent accesses within a sliding window to estimate the probability of future document accesses. In addition, we take into account the variable sizes of documents. Simulations with real-life web access data are conducted to evaluate the performance. SSW-LFU outperformed other algorithms in hit ratio and had comparable byte hit ratio when compared with other algorithms such as LFU, LRU, Size, LRU-Min, etc.