We used the Composite Burn Indices sampled in the field to test performance of radiometric measures as estimators of burn severity. Two 1994 fires occurring at Glacier National Park, Montana, were investigated. Indices incorporated band ratios and multi-temporal differencing derived from the Landsat Thematic Mapper, including: 1) post-fire band 7 reflectance; 2) pre- and post-fire differenced band 7 reflectance; 3) differenced Normalized Difference Vegetation Index (NDVI), and 4) differenced Normalized Burn Ratio (NBR), a new index formulated from TM band 7 and band 4 reflectance. Seasonal effects also were tested, with indices obtained from spring and late summer data, for a total of eight models. Evaluation of performance considered amplitude of index response, direct correlation to field-rated burn severity, and visual characteristics of derived images. NBR differenced from early in the growing season was judged the most effective measure of burn severity. Greater amplitude of response and correlation to field-rated burn severity led to better contrast, broader range of severity levels, and sharper delineation of burn perimeters. NDVI differenced late in the growing season exhibited the poorest resolution of burn severity overall. In general, results were improved for all indices when derived early in the growing season. Seasonal differences, however, were influenced by phenology and not strictly by date. Spring scenes performed best at low elevations, while late summer results were stronger at high elevations, where snow melt and plant growth was delayed by at least two months. To facilitate assessment of burn conditions through mapping and summary tabulation, differenced NBR was partitioned into 7 ordinal levels (including meadow, forest, and unburned classes) which exhibited strong relationships to burn severity rated on field plots. Advantages differenced NBR include: 1) a standard, transferable measure with reduced influence from image- or observer-dependent qualities; 2) utility on large, remote burns and potential to compare multiple burns regionally or nationally; 3) direct correlation to field-estimated fire effects and other quantitative environmental variables; and 4) flexibility to apply the measure as a continuous or discrete variable in modeling, research and management.