Abstract: By analysis of web server logs in e-commerce to extract the information of web consumer's products browsing behavior, the weight of a product of web consumer's current preference has been estimated by means of consumer's browsing frequency, browsing time, the number of paths and links depth. The theory of bidirectional association rules was combined with the idea of Apriori to find out interdependent products with to discover consumer's product preference path, candidate products of consumer preferred have been found based on consumer's current browsing behavior, and then each preference degree of consumer to those products has been calculated respectively. Finally, the web server logs of a self-developed e-commerce web site has been used in the simulation to find out web consumers tour preference. Experimental results shows that the recommendation accuracy of preference mining method based on bidirectional association rules has improved greatly and the coverage has been expanded compared with preference mining method based on association rules when the experimental condition is equal.