The avoidance of collisions between ships in encounter situations is crucial to maritime traffic safety. Most research on maritime collision avoidance has focused on planning a safe path by which to avoid approaching ships in accordance with the requirements laid out in the International Regulations for Preventing Collision at Sea (COLREGs). The resulting solution provides reference for the navigator in planning movements to avoid collisions. Nonetheless, specific anti-collision actions are generally based on the experience of the navigator. This study differed from existing works in that we sought to derive collision avoidance behavior from the trajectory data of actual ships. The Automatic Identification System (AIS) network makes it possible to collect an enormous volume of trajectory data and investigate real-world ship behavior. Unfortunately trajectory data that introduces uncertainty can hinder behavior mining for collision avoidance. Irregular and/or asynchronous location sampling can lead to situations in which the movement of a ship does not necessarily follow a given trajectory, even if its movement behavior is similar to that of other ships. In this study, we developed a framework to decipher the anti-collision behavior of ships in encounter situations from a large database of trajectory data collected by AIS network, and to present this behavior in the form of anti-collision patterns. A prototype of the proposed framework was built to enable pattern analysis and visualization functions, thereby providing a deeper understanding of collision avoidance behavior in maritime traffic. The proposed framework is applicable to the development of pattern-aware collision avoidance systems aimed at improving maritime traffic safety.