The traditional approach to detect mental state mainly depends on questionnaires completed by the subjects. The accuracy of answer is affected by many factors including subject's knowledge, the ability of memory, and the authenticity of answer. In order to improve the accuracy, we propose to utilize web behaviors to identify mental state instead of questionnaires. Currently, the web plays a key role in most people's lives, and we may acquire web user's access history more easily and accurately. Thus we are able to make the identification based on the objective observations. In this paper, we investigate various web behaviors which relate to mental health, and discuss the possibility to use web behaviors to diagnose related mental health under different circumstances.