Traditional agricultural information systems lack availability because of ignoring user preferences. This paper proposes a user interest modeling method which combines auto-modeling with artificial modeling. The value of user interest measure can be obtained by counting the frequency of information accesses. The statistics method based on the heat of information accesses is adopted to perform cluster agricultural information push. The result of experiment proved the user interest modeling method can express a user’s interest in agricultural information better. The intelligently and personalization of the system are improved.