A new walking pattern classification method is proposed for a 5-link 7-DOF biped robot walking on an uneven floor. This method extracts the patterns in the current floor position of the stance foot and the transitioning floor conditions of the swing foot during locomotion. When a global path composed of stairs, obstacles, etc., and certain walking parameters, such as the speed of walking and the total walking time, are put into the system, the guidance controller unit determines the trajectory of the footsteps in terms of step patterns by using a genetic algorithm-based optimization technique while ensuring the biped's stability criterion. A demonstration of the biped with different pattern classes was realized by a dynamic simulator. © 2011 International Symposium on Artificial Life and Robotics (ISAROB).