Abstract We present a new method to obtain the nonlinear zero drift of DZW gravimeter by using gravity observation data. The main algorithm of this method is a Bayesian formula. The principle is to take the continuity of zero drift as an a priori constraint, consider the deviation caused by the floating of gravity observation data, and solve the mean and variance of drift at each sampling point as hyper-parameters. The algorithm in this paper can be used repeatedly. With the increase of evaluation time, the hyper-parameters will tend to the convergence limit, so as to obtain the optimal solution of the algorithm. Through data verification, the accurate nonlinear zero drift curve can be obtained. During zero drift correction, the variance of gravity observation value can be obtained which is 1.595 53×10-10 [μGal]2 . The algorithm in the paper ensures the measurement accuracy of DZW gravimeter.