Analysis of network user behavior inside enterprise intranet becomes more and more crucial for enterprise network management and business analysis of internet applications. In this paper we focus on the web surfing analysis to capture the users' behavior. We examine the web surfing classification data both on individual level and network level. 12 classes of web pages are taken into consideration and the data of more than 10 thousand of users from 188 enterprises over an epoch of 15 days are collected. For every subject under examination, the web surfing frequency of the 12 classes forms a 12-dimensional vector. We cluster all the vectors on different levels and the users and enterprises are classified respectively. To evaluate the stability of users' behavior, we introduce a stability measure for web surfing behavior. We calculate this measure for all the users and the results are analyzed. The web surfing behavior classification and behavior stability analysis are very useful for the enterprise manager to find out inappropriate use of the enterprise network. The analysis of web surfing behavior of the whole enterprise can also be used by the web sites to find the web surfing patterns of different kind of enterprises.