Relative position errors among mirrors in a three-mirror reflective system can seriously degrade the performance of the whole space optical remote sensing system, so we have to calibrate these errors in orbit. A new wavefront sensorless calibrating method that integrates stochastic parallel gradient descent control algorithm, phase-diversity method and sensitive matrix inversion algorithm is proposed. This method retains the high resolution of sensitive matrix inversion algorithm and largely increases the dynamic range of stochastic parallel gradient descent control algorithm. Additionally, it does not need wavefront sensors to measure the wavefront error, thus simplifies the whole system, making it more feasible. Through computer simulation of a typical high-resolution space optical system in orbit and calibration of its relative position errors of the three mirrors using this integrated method, significantly improved calibrating range and high calibration precision are demonstrated, thus the merits of the new method are proved.