In a distributed fiber optic temperature sensing system, the intensity of Raman Stokes backscattering light serving as reference light increases with the increase of temperature, leading to measurement errors in the system. A novel method of dynamically calibrating Raman Stokes backscattering light intensity is proposed to improve temperature accuracy for distributed fiber optic temperature sensors. According to the real-time Stokes intensity distribution in the reference fiber, Stokes intensity curve of the whole fiber at a reference temperature is simulated, and the temperature response of Stokes light is corrected. The ratio of Raman anti-Stokes light intensity to the calculated Stokes light intensity is used to demodulate temperature along the fiber. The experimental results indicate that the temperature accuracy of the distributed optical fiber temperature sensor system after making the Stokes optical dynamic calibration is increased up to 4.3 ℃ compared with that from the conventional method. And the accuracy of temperature measurement is improved by 8.9 ℃ when combined with Rayleigh noise suppression method. This study provides a new solution for a distributed fiber optic temperature sensor system to monitor high temperature environment temperature.