Various heating processes are used in a steel works. However, emissivity compensation is still one of the most difficult problems when radiation thermometers are applied to temperature measurements of heated sheets in steel manufacturing processes. The authors propose a new technique using spectral information of radiation from targets and multivariate analysis, such as principal component analysis (PCA) or partial least squares (PLS) analysis, etc. First, spectral radiation from a target is measured by a hyperspectral camera, and the scores of a predetermined spectral component are calculated as inner products of the measured spectral radiation and the predetermined spectral component. Second, the spectral component is predetermined so that its scores change with temperature and are minimally affected by the deviation of spectral emissivity. For example, the component can be determined by PCA under the condition that it is perpendicular to the deviation of spectral emissivity, which is evaluated in advance. Finally, temperatures are calculated from the scores and a calibration curve relating the scores to temperatures. The calibration curve is determined in advance from the measured spectral radiations from a blackbody furnace and the scores of the component at several temperatures. The authors developed a radiation thermometer using a hyperspectral camera and installed the camera at an annealing furnace in the stainless steel manufacturing process. As a result, it was found that the standard deviation and the maximum error of the developed radiation thermometer from the values measured by thermocouples were less than those of an ordinal single-wave thermometer.