Abstract: The PSP pyranometer is used as a reference pyranometer and a reliable in strument in China. Before 2006 it is once used as national diffused reference to measure the reference diffused irradiance, the value of which is used in the sum composition method to calculate the reference global irradiance. The probability distribution of the nighttime thermal offsets within the pyranometers demonstrates near normal distribution. The offset of PSP type pyranometers is larger than that of 8-48 and CMP22 type. Finding the characteristics of the offset within different types of pyranometers and doing the calibration will improve the accuracy of diffused irradiance. The pyranometer calibration is done underclear sky. The negative thermal offset of the pyranometer causes that the sensitivity is underestimated. By the experiment and study on the thermal offset correction method for different type pyranometer, five correction methods are compared. The net longwave sensitivity method: Calculating the net longwave sensitivity of the pyranometer using longwave blackbody radiation, which is used in the sum composition method outdoors to correct the thermal offset of the pyranometer to reduce the offset effects. The differential algorithm with the reference global irradiance: Comparing it with the reference value derived from the sum composition method using the absolute cavity radiometer and CMP22 type secondary standard pyranometer with shading. The net longwave equation method: Analyzing the data of pyranometers and PIR type pyrgeometers measured at night to correct thermal offset. The cover method: Covering the dome of pyranometers in daytime to measure the offset of pyranometer directly. Using the voltage output of pyranometer in the nighttime as the thermal offset: Calculating the average voltage (night) when solar is 7° lower than horizon and transform it to irradiance as the thermal offset. When using the sum composition method to calibrate PSP pyranometer, the correction results of the net longwave sensitivity method is better than other methods. The result of the cover method is more close to the net longwave sensitivity method, the correction effects using the night voltage output method is not good. Drawing the daytime offset curves separately using the five offset correction methods, the net longwave equation method and the net longwave sensitivity method is more close to the reference value. But these methods are only confined to the shaded pyranometers, because shading makes pyranometer performs more like in the night. For unshading pyranometers, because of the effects of direct irradiance, the five correction methods are distinctly different with the true value. When using CMP22 and 8-48 pyranometers with shading in the sum composition method to test the reference diffused irradiance, the thermal offset could be ignored. But the CMP22 or 8-48 pyranometer should be calibrated by "shade/unshade" method to reduce the effects of offset for sensitivity.When the net longwave sensitivity method or the cover method is used in the sum composition method, the accuracy of PSP's sensitivity could be improved 1%.