The invention provides a magnetometer calibration method based on a Levenberg-Marquardt algorithm, and belongs to the technical field of the optimization theory, numerical analysis and non-linear regression. The method comprises the steps: carrying out the non-linear fitting calibration through the introduction of the Levenberg-Marquardt algorithm, thereby preventing an individual piece of data affected by an accidental error from affecting the calibration, avoiding a condition that a conventional Newton-Gauss method may just obtain poorer local minimum estimation during non-linear fitting. Moreover, the method employs the characteristic that the data structure of magnetometer fitting is simpler, and is better in global secondary convergence. The Levenberg-Marquardt algorithm is stronger in global convergence. Although the Levenberg-Marquardt algorithm is provided with a better initial solution value during the processing of the original data of the magnetometer in different sections and is helpful for the convergence speed, finally the algorithm can achieve the better fitting of the input data.