A focused crawler is designed to traverse the Web to gather documents on a specific topic. It is not an easy task to predict which links lead to good pages. In this paper, we present a new approach for prediction of the important links to relevant pages based on a learned user model. In particular, we first collect pages that a user visits during a learning session, where the user browses the Web and specifically marks which pages she is interested in. We then examine the semantic content of these pages to construct a concept graph, which is used to learn the dominant content and link structure leading to target pages using a Hidden Markov Model (HMM). Experiments show that with learned HMM from a user's browsing, the crawling performs better than Best-First strategy.