Aiming at the measurement errors caused by nonorthogonality, inconsistent sensitivity, zerooffset of the triaxis fluxgate sensor and the position misalignment of two fluxgate sensors composed of the gradiometer, the error model is established in this paper. Based on the invariance principle of total geomagnetic field vector modulus in a short time, the parameter identification of the first fluxgate sensor is performed with linear least squares algorithm; based on the principle of the minimum F norm of the 3 component difference after calibration, the parameter identification of the second fluxgate sensor and the relative position between two fluxgate sensors are conducted with the multiple linear regression method. The experiment results show that the proposed method can reduce the maximum total magnetic field error of two fluxgate magnetometers from 1 194.4 nT to 30.0 nT, and the maximum output error of the three component gradiometer from 529.1 nT to 13.4 nT. The performance of the gradiometer is effectively improved.