An on-line three-dimensional temperature measurement experiment was carried out in a gas-fired pilot tubular furnace. Four flame image detectors were utilized to obtain two (red and green) monochromatic radiation intensity distributions, which can be calculated by the DRESOR method based on the radiation image processing technology. Then a revised Tikhonov regularization method was developed to reconstruct three-dimensional temperature distributions from the green monochromatic radiative intensity. Meanwhile, a Newton method combined with a least-squares method was used to simultaneously reconstruct radiative properties from the red one. The two calculation procedures were performed alternately, forming an iterative algorithm to a simultaneous reconstruction of temperature and radiative properties. The reconstructed temperatures agreed well with those measured by thermocouples for different cases with different calorific values and components of gas. The largest relative error was less than 3%, which validated the effectiveness and accuracy of this reconstruction algorithm. Moreover, the nonuniform radiative properties for the flame and nonflame regions were determined to improve the accuracy of temperature measurement by a rigorous comparison test. Finally a set of reasonable fixed radiative properties for the media and walls was chosen for the on-line detection of temperature. The visualized temperatures obtained by the present method agreed reasonably with those measured by thermocouples for all cases, with the largest relative error less than 5%. The present method based on radiation image processing technology is reliable for on-line temperature measurement and shows a good accuracy for its application in the combustion industry.