The effectiveness of an adaptive acceleration method is studied for the inner iterations in some neutron diffusion codes. The acceleration method can be easily incorporated in the process of the general first-order stationary linear iterations. The effectiveness is shown theoretically in the case when the iteration matrix is nonnegative definite. The numerical results of its applications to the successive over-relaxation and alternating direction implicit iterations are presented. It is shown to work effectively even when the fixed parameter of the iteration is not chosen optimally. Some variants of the acceleration method are also given.