A vision-based fuzzy obstacle avoidance method is designed and implemented on a humanoid robot so that it can avoid obstacles successfully and arrive at the terminal area effectively. A humanoid robot with 23 degrees of freedom is implemented so that it can execute six basic walking motions. One vision system and one electronic compass are installed on the robot to obtain the environment information so that it can obtain the environment information to be an autonomous mobile robot. In order to avoid obstacle successfully, the minimal distance between the robot and the obstacles in the moving direction measured from the captured image of the vision system is considered as a dangerous factor in the moving direction. In order to attend at the terminal area effectively, the angle difference between the goal direction and the moving direction of the robot measured from the electronic compass is considered as a helpful factor in the moving direction. The dangerous factor and the helpful factor are considered to be two inputs of the proposed fuzzy system to evaluate the feasibility of each motion so that one of the six motions with a highest value is selected to be the next motion in every decision. Some simulation results in four different environments by placing different number of obstacles and one practical experiment of a difficult environment are presented to illustrate the effectiveness of the proposed method.