Abstract: Hot spot is an important problem in load balancing research of Web cache cluster. The existence of hot spot could easily induce access skew and load imbalance. After analyzing features of hot spot, a model to predict hot spot access frequency is built based on grey prediction. The proposed procedure of new hot spot relief strategy can be described as follows: set a threshold, distinguish object, whose predicted access frequency exceeds the threshold, as hot spot, push hot spot into all nodes, implement redundant store for hot spot, and finally achieve the load balance in Web cache cluster. Compared with other strategies, grey prediction based hot spot relief strategy heightens the performance of Web cache cluster.