Collision of ships is a typical maritime accident that usually happened at a low rate but with disastrous consequence. Human elements play a very important role in collision avoidance. It is reported that human elements were involved in more than 90% of collision accidents. The most important thing for collision avoidance is to make sure that the human being in collision avoidance system is reliable. In this paper, a method called technique for human reliability analysis in ship collision avoidance or THRASCA is presented, which is purposely designed for assessing the human reliability in collision avoidance of ships quantitatively. The analyzing procedure is consisted of seven steps: determining the scope of analysis, initial task analysis, identifying the human elements, expressing the human errors, identifying the performance shaping factors, quantifying the human reliability and expressing the human reliability. Since the method is simple in expression, it could also act as an active aid for those who are in charge of maritime safety affairs. The main difficulties met in designing the method are lack of human element data, lack of proper human error model, the vagueness of the relationship between human error and its shaping factors. To overcome all these difficulties, accident statistics, written accident reports, expert judgment and simulator experiment are taken as the data sources of human element; the techniques such as gray relational assessment, fuzzy logic and rule discovery are used as tools in identifying human elements and their relationships; the human error model are established on the recognition process of collision avoidance as sensing, decision making and action.