This paper considers the application of pattern recognition techniques in modern computer games. Towards the problem of realizing more life-like behavior for artificial game characters, we record the network traffic of online multiplayer games. Dealing with a soccer game, we cluster these data and train HMMs in order to achieve fast and robust recognition of behaviors and actions in the virtual game world. Experimental results indicate that pattern recognition and machine learning provide an auspicious avenue towards more convincing artificial characters.