Fire can be used not only to protect forest ecosystems but also as a restoration tool. However, a successful application of fire requires knowing in advance the potential fire behavior (e.g., rate of spread and intensity). Without this information adverse effects could result when fire is prescribed in forest ecosystems. Fire behavior can be simulated through the use of many of the systems that currently exist. However, since all of them require reliable spatially referenced data, their use could represent a more complicated problem. Fuels characterization and their spatial distribution are critical factors to simulate fire behavior. In an attempt to simplify the fuels characterization, the "fuel model” concept was created. This concept allows the categorization of areas into classes of potential fire behavior. However, the definition of the spatial distribution of such classes has represented one of the more complicated challenges facing forest fire scientists. Nevertheless, new technologies, such as remote sensing, GIS, and computer modeling, permit alternative strategies for classifying a given area into its corresponding fuel models. In this sense, this paper presents one of the first attempts to classify a forest region into fuel model classes, as a support to simulating fire behavior in Mexican forest ecosystems. A traditional forest inventory, complemented with fuels surveys, was carried out in 1998 in Chihuahua, Mexico. A total of 554 plots, randomly distributed, were sampled in a 1400 ha area. Using the quantities and proportions of 1-hr, 10hr, and 100-hr fuel loads, each sample plot was classified into its corresponding fuel model. Ordinary kriging (a geostatistical technique) was used, through a geographical information system, to interpolate the fuel model values of the sample plots. The resulting layer was used to support a supervised classification of a Landsat TM5 image, and to enhance an unsupervised classification of the same image. The corresponding accuracy assessment of those classifications resulted in a slightly higher accuracy for the supervised classification. These results can be extrapolated to other regions similar to the study area, where fuels inventory may not exist. Although the results were acceptable, further investigation is suggested, not only to validate this methodology under different forest ecosystems, but also to use other geostatistical interpolation techniques (e.g., co-kriging) to include more parameters for the classification of a region into fuel model classes.