Power sensor nonlinearity contributes significantly to the increase of the microwave power measurement uncertainty. Current methods for sensor calibration do not provide correction of the results, due to the sensor nonlinearity. The paper describes an improved method and an automated measuring system based on it for power sensor calibration that enables the correction of the results, taking into account amplitude and frequency dependent nonlinearity. The novel calibration method is based on the application of high-linearity thermistor power transfer standard. The power sensor calibration process is fully automated and improved by applying the automated measuring system, PC and VEEpro software. Performed calculation of the calibration measurement uncertainty and the analysis of power measurement uncertainty budget indicate that the new calibration method allows the correction of sensor nonlinearity and contributes to significant reduction of the microwave power measurement uncertainty, ranging from 15,8% to 40,5%. Experimental results and validation confirm the applicability of the improved calibration method and prove the existence of significant dependence of sensor nonlinearity on the power level, but also on the frequency of the measured microwave power.