There have been many studies of load balancing in computer systems, for which various algorithms have been proposed. The load balancing methods to date can largely be divided into two types. In the first method, the load information is exchanged among the nodes, and the loads of the nodes are balanced by transferring the request calls. In the second method, the load balancing scheduler in the system monitors the load state in each node, and allocates request calls so that the loads of the nodes are balanced. In contrast to those methods, this paper proposes a load balancing method in a distributed autonomous cooperative system, where the nodes operate asynchronously according to their own decisions. In the proposed method the scheduler is not necessary, and load information is not exchanged or request calls transferred among the nodes. As a result of evaluation by simulation, it is seen that the proposed load balancing method can be applied to a system composed of nodes with uniform performance, as well as to a system composed of nodes with nonuniform performance. The following results were also obtained. (1) In the system composed of nodes with uniform performance, the average response time is less when the load is high, compared to the load balancing methods used to date. (2) In the system composed of nodes with nonuniform performance, the load can be balanced to reduce the average response time not only for the case where only CPU processing is considered in the node, but also for the case where both CPU processing and IO processing are considered.