A procedure is described for removing two error vectors from the approximation to the solution obtained iteratively. Such a scheme for accelerating convergence can reduce the computation cost of solving neutronics problems as is demonstrated for finite difference diffusion theory.