Fires spread heterogeneously across landscapes resulting in patches of varying burn severity levels. Burn severity, caused by changes in canopy cover, biomass, and moisture, can be detected and mapped using satellite data. The 2001 Leroux fire on the Coconino National Forest, Arizona was studied using existing pre-burn measurements on overstory, understory, and fuels in twenty-four 0.10 ha permanent plots. Following the fire, plots were re-measured. Landsat 7 ETM+ imagery and the Normalized Burn Ratio were used to map the fire into four severity levels immediately following the fire (July 2001) and one year after the fire (June 2002). Overall KHATs were 0.67 for the 2001 image and 0.62 for the 2002 image, showing that both were statistically better than random. However, there was no statistical difference between the 2001 burn map and the 2002 burn map. Individual burn severity classes were classified correctly 71-83% of the time in 2001, and 69-80% of the time in 2002. The most accurate classes were moderate severity and high severity in 2001 and 2002, respectively. Confusion was greatest between unburned and low severity classes. These errors occurred 7.6% of the time in 2001 and 10.9% of the time in 2002. Errors are attributed to abundant re-growth of vegetation in 2001 and extreme drought in 2002. We analyzed pre- and post-fire plot measurements according to their imagery classification. According to imagery classification, severely burned areas were predominantly P. ponderosa stands as calculated by density and basal area. Tree density and fuel accumulation also influenced severity levels. However, when post-fire plot mortality was summarized by severity levels as determined by the imagery, overall mortality was not greatest in severely burned areas. Therefore, mortality alone may not be a good indicator of burn severity level. This result led us to develop a more comprehensive definition of burn severity using all measurements on soils, fuels, understory, and other tree measurements such as char and scorch.