According to the TAM （Three Axis Magnetometer） calibration problem in the unknown noisy environment， a novel unknown measurement noise compensation method is proposed to adaptively calibrate vector magnetic sensors. On the basis of a more complete modeling of the calibration， several shortcomings of the traditional calibration method are pointed out when the error parameter is large. The measurement data matrix and the noise matrix in the unbiased cost function are separated using the convolution formula.The unknown noise amplitude in the noise matrix is obtained by constructing a first-order Schur complement approximation after blocking. Through this method， the unknown noise in the measurement data is completely removed.Both simulation and physical experiments demonstrate that the accuracy of the calibrated vector magnetic sensor is significantly improved after noise compensation compared to before compensation， with the advantage becoming more pronounced as noise increases.