A model is designed and implemented for mining user frequent paths from Web log. An important basis for webpage clustering analysis is proposed, as well as the concepts of page value and jump preference degree. Then we build the page value model. The distances of page value are calculated out from the Page Value-User Matrix, then the value-equal page set is obtained according to the distances. After that, the set is transformed to binomial frequent path set. Finally the user frequent path set can be obtained by applying an adaptive merging algorithm. Experimental results show the model has better accuracy with high efficiency.