Due to the influence of crosstalk, AC-Stark effect and other factors, the spatial position ( three-dimensional coordinates and sensitive axis pointing) and gain coefficient of the SERF atomic magnetometer are deviated, which directly affects the accuracy of magnetic source positioning. To address the above problems, an adaptive calibration method of SERF atomic magnetometer parameters is proposed. Based on the magnetic dipole model, a calibration device is designed. It consists of 24 precision-machined circular coils, which are used to apply the calibration magnetic source. An improved adaptive elite genetic algorithm is proposed to simultaneously calibrate the relevant parameters of the magnetometer. Experimental results show that the average correlation coefficient between the actual magnetic field curve and the theoretical magnetic field curve fitted by the algorithm is 99. 55% , the x-axis coordinate value drift is the most obvious, the average absolute deviation is 2. 63 mm, and the average absolute deviation of the sensitive axis is 8. 21 °. Results show that the sensor parameters need to be accurately measured before the magnetic source can be located. The proposed calibration method has certain reference significance for improving the positioning accuracy of the magnetic source.