In this paper, a law similar to that of the Wien's law, but to determine the optimal wavelengths in a mono-spectral and a bi-spectral methods (without the Wien's approximation, i.e. for a Planck law) used for temperature measurement of surfaces exhibiting non-uniform emissivity, and a more general methodology (based on the ordinary least squares method) to obtain the optimal wavelengths selection in a multi-spectral method is presented. The goal consists of minimizing the standard deviation of the estimated temperature (optimal design experiment). For the multi-spectral method, two cases will be treated: optimal global, and optimal constrained (to the spectral range of the detector, for example) wavelengths selection are presented. The estimated temperature results obtained by different models taking into account a secondorder polynomial transfer function and including the emissivity variations and for different number of parameters and wavelengths are compared. Different selection criteria are presented. These points are treated from theoretical, numerical and experimental points of view.