Abstract: The amount of information on web is increasing at an alarming rate. It is an urgent need to find tools to automatically obtain, extract and filter information from web, from hundreds of millions of pages to find the content in need, to find related patterns and associations. Markov model predicts the user's next link, only from the user's browser start page, which does not involve the real interest of the user. In this paper, HMM-based prediction of the web browser path is presented. First of all, according to known sequences to determine the browser type of user. As can be seen for the browser with a very short sequence length, the accuracy of the forecasts is lower than the Markov model. This is due to the short sequence length, the system can access only limited information to make judgement, with more classification of user errors as more likely. However, with a gradual increase in sequence length, the system may capture more and more user's browsing information to reflect user's interest and to increase the accuracy of prediction. When the sequence length is greater than or equal to 8, the forecasting accuracy rate reaches 80%.