A Bayesian network classifier-based algorithm was applied to map the burned area (BA) in the North American boreal region using the 0.05° ( ~ 5\nbspkm) Advanced Very High Resolution Radiometer (AVHRR) Long-Term Data Record (LTDR) data version 3 time series. The results showed an overall good agreement compared to reference maps ( slope = 0.62; R2 = 0.75). The study site was divided into six sub-regions, where south-western Canada performed the worst ( slope = 0.25; R2 = 0.47). The algorithm achieved good results as long as a year with high fire incidence was employed to train the Bayesian network, and the vegetation response to fire remained consistent across the region. Years with higher fire activity and larger fires, which were easier to detect at the LTDR spatial scale, matched the reference maps better. The LTDR postfire signal remained detectable for 6-9 years, extending opportunities to map the full fire extent with Landsat Thematic Mapper (TM). For fires larger than 1000\nbspkm2, Landsat TM mapped 99%, whereas LTDR caught 69% of the reference BA reported. Landsat TM took four satellite overpasses (2 months) to map these large fires, and in some cases even until the following year, but LTDR detected them within days. Thus, results suggest that LTDR could be used to trigger the search for fires and then map their perimeter with Landsat TM. This study demonstrates an LTDR BA algorithm that could be extrapolated to other boreal regions using a similar methodology, although reference fire perimeters would be needed to train the Bayesian classifier and its thresholds.