A novel knowledge extraction method from autonomous behavior of multiple mobile robots by an artificial life approach is proposed in this paper. The knowledge is expressed by if-then rules and we employ a neural network for the knowledge extraction. The proposed method has the following features: 1)The structure of knowledge extraction neural network and the learning algorithm are simple; 2)Understanding and modification of the extracted knowledge are easy because weights in the knowledge extraction network directly represent the antecedents and the consequents of the if-then rules; 3)The network itself has an ability of inference using the extracted knowledge. We used a lot of autonomous mobile robots in various environments. Each robot has to avoid the obstacles to get to the goal and the local behavior is extracted and integrated in the knowledge extraction neural network as global knowledge. We confirmed the validity of the proposed method by computer simulations.