///The authors analyze the ultra large-scale sparse linear equations developed in the course of the 3-D geoelectric field forward modeling and describe emphatically Lanczos iterative process and the algorithm principle for solving such equations in this paper. When there exist big difference in electrical properties among the subsurface media which will makes condition number of the coefficient matrix is big, the authors make some changes to the algorithm and use incompletely Cholesky decomposition method to carry on preconditioned processing. The coefficient matrix with such improvement is approximately a unit matrix and the improved Lanczos algorithm will improve numerical stability and convergence of iterative procedures. The algorithm is an efficient method of 3-D geoelectric field forward modeling, and makes a firm foundation base for 3-D geoelectric inversion.