Queries against the large size XML database are often exploratory and users often find their queries return too many answers, this paper proposed a user preference-based top-k ranking approach to deal with this “information overload” problem. We first presented a user preference model which can embody both the partial relations and the interest degree of preferences. And then, the elements orders of XML database are created by considering the user preferences and consequently the representative orders are computed by using the clustering algorithm during the offline step. Finally, based on representative orders selected in the offline time, the the Top-k result elements are selected by using TA algorithm during the online processing step. The efficiency and effectiveness are demonstrated by the experiments.