Research on the calibration model of MEMS accelerometer is an important method to improve the accuracy of MEMS accelerometer. An accelerometer calibration model based on improved Levenberg-Marquardt algorithm is proposed in this paper. Based on the static multi-position inversion method for calibration, according to the error to establish a mathematical model, namely the nonlinear least square minimization problem. Because the original Levenberg-Marquardt algorithm is slow to decrease and has a large amount of computation in solving the optimal estimate, the iterative times of obtaining the optimal estimate are reduced by making full use of the calculation results of each iteration of the algorithm to set the step factor, and the convergence of the improved algorithm is proved theoretically. In view of the problem that the value deviates from the real value after calibration, it is proposed to use the sensor state information to improve the calibration model, and the numerical experiments using the improved calibration model have good results.