Currently, the implementation of WEB caching is mostly based on traditional cache updating algorithms. However, due to the diversity of the WEB traffic pattern, the traditional algorithms for cache updating can not be used in WEB environment effectively. In this paper, an optimized model to the problem is presented. The analytic result shows that the key issue for the cache updating algorithms is how the algorithm suits the WEB traffic pattern properly. Based on the Poisson arrival model, a new cache policy, Least Normalized Cost (LNC), is proposed. In addition to the consideration of the average reference time duration, the recently passed time, the size of the WEB file and the cost per unit of file, the dynamic characteristic of WEB access rate is also taken into account. The trace driven simulation shows that the performance of the algorithm LNC is better than that of the existing algorithms proposed in the literature.