This paper provides a definitive solution for solving the three-axis magnetometer calibration problem by means of the Procrustes analyze. The non-orthogonal Procrustes analysis is a set of mathematical least square tools to directly and efficiently estimate the optimal similarity transformations among two point clouds. The explicit solution of calibration parameters is equivalent to the translating, rotating and scaling in the Procrustes problem. This direct solution greatly facilitates the real-time application, since the computation time consumption is determined directly by the number of sample points now. Simulated and experiment results are presented, supporting the accurate and robustness of the proposed solution.