The methods of temperature measurement power exponent curve fitting and cubic spline interpolation radiation calibration of an infrared thermal imager are analyzed. The radiation calibration of a typical middle wave (MW) infrared thermal imager is carried out by using a high accuracy temperature control surface source blackbody to obtain the correct relationship between the blackbody temperature and the output voltage signal. The power exponent the least square fitting and cubic spline interpolation are performed on the calibration experimental data, and the temperature measurement error of the two methods are compared. The result shows that the temperature measurement errors of the two fitting methods decrease gradually with increasing of fixed punctuations. The power exponent least square method can obtain higher temperature measurement accuracy when three temperature points are used in the range of temperature measurement. In the range of temperature measurement, the 3-points punctuation with uniform distribution of temperature interval can obtain higher accuracy temperature measurement than that of uniform distribution of gray interval. The temperature measurement accuracy of this calibration method can reach 0.1 K, which is an ideal, fast and accurate temperature measurement method for thermal imager.