Accelerometer is one of the fundamental measurement units of inertial navigation system. It is difficult to meet the precision requirement for low-cost accelerometer due to the manufacturing process and all kinds of sensor errors. Calibration for accelerometer is essential before being used. Therefore, an accelerometer self-calibration algorithm based on maximum likelihood estimation was proposed. The sensor errors model was established by taking comprehensive consideration of zero bias, scale errors, non-orthogonal errors, installation errors and measurement noise of the accelerometer, based on which the calibration problem of accelerometer was transformed into maximum likelihood estimation problem of calibration parameters. The self-calibration algorithm based on maximum likelihood estimation was tested by both numerical simulation and real data experiment. The result shows the maximum likelihood estimation algorithm has a high precision of parameters estimation and can calibrate the errors caused by factors mentioned above effectively.