Web usage mining (WUM) is the type of Web mining activity that involves the automatic discovery of user access patterns from huge Web access logs. In this study, we analyze deeply generalized suffix tree data structure in WUM situations and explain in detail the reasons why a linear-time traversal on the generalized suffix tree can obtain frequent reference sequences. The key point is that due to the special nature of transactions, for each internal node v, the total number of leaves in the sub-tree of v is exactly the number of distinct (navigation-content) transaction identifiers that appear at the leaves in the sub-tree of v. After that, with the help of generalized suffix tree, an algorithm on mining maximal reference sequences is proposed. Experimental results indicate that our approach is feasible and has good scalability.