Resumo. Models relating spectral answers with biophysical parameters aim estimate variables, such as wood volume, without the necessity of frequent field measurements. The objective was to develop models to estimatewood volume by Landsat 5 TM images, supported by regional forest inventory data. The image was georeferenced and converted to spectral reflectance. After, the images-index NDVI (Normalized Difference Vegetation Index) and SR (Simple Ratio) was generated. The reflectance values of both spectral bands (TM1, TM2, TM3 e TM4 ) and indices (NDVI and SR) was related with the wood volume. The biggest correlation with volume was with the NDVI and SR indices. The variables selection was made by Stepwise method, which returned three regression models as significant to explain the variation in volume. Finally,the best fitted model was selected (volume = -830,95 + 46,05 × (SR) + 107,47 × (TM2)), which was applied on the Landsat image where the pixels had started to represent the estimated volume in m³/ha on the Eucalyptus sp. production units. This model, significant at 95 % confidence level, explains 68 % of the wood volume variation.