As channel selection acts an important influence on calibration accuracies in the calibration of imaging spectrometers by transfer radiometer, herein MATLAB language was adopted to compile a merit function, which was used as the evaluation standard of inversion accuracy of the spectral curve of a blackbody at a temperature of 3000k. The MATLAB genetic algorithm and the pattern search algorithm toolbox were employed to solute the minimal value of function that was the optimal spectral channel distribution of the transfer radiometer. Finally, influence factors of the number of spectral channels, including the shifting of center wavelength, the broadening of bandwidth and measurement system error, were analyzed. The results show that the spectral inversion accuracy of is 0.12%, when the shifting of center wavelength does not exceed 0.2 nm, the broadening of bandwidth is not more than 0.02 nm, and the system error is less than 0.1%.This study can guide the design of transfer radiometers and has a great significance to improve the calibration accuracy of earth imaging spectrometers.