This work presents a numerical and experimental study in a laboratorial rig to measure local flame temperature during the biomass combustion. Most of practical situations of biomass combustion is a transient and non-stable phenomenon. A spectral analysis method in the visible spectrum range was used for measurement of local temperatures in a sugarcane bagasse flame. The two-color method considering spectral flame emissivity variation joint with an inverse numerical procedure is used to simultaneous estimation of flame temperature and spectral emissivity. Also, a sequential iterative numerical procedure is tested. It includes the application of two-color method considering grey emissivity between two near spectral intensity data in a first step. In a second step and iterative procedure considering spectral emissivity variation along whole spectral range is applied. To apply the numerical methods was used several spectral intervals in the visible range. In inverse estimation, the LevenbergMarquardt method is used and a polynomial and a cosine models for estimation of coefficients were tested. Six test of biomass combustion were made focusing the sensor to the reaction zone, and spectral data collected. The flame temperature estimated for the six spectral data were closed to 1530 to 1540 K. Also, the spectral emissivity is closed in all sixcombustion test. Instantaneous bagasse flame temperature in the range of 1400 K to 1500 K in the reactive combustion zone were found. An image processing method was used to measure the temperature in the same reactive combustion zone and the comparison show reasonable agreement. The results in this work also are in reasonable agreement with results in literature.