We evaluated agreement in the location and occurrence of 20th century fires recorded in digital fire atlases with those inferred from fire scars that we collected systematically at one site in Idaho and from existing fire-scar reconstructions at four sites in Washington. Fire perimeters were similar for two of three 20th century fires in Idaho (1924 and 1986). Overall spatial agreement was best in 1924 (producer's accuracy = 94% and 68% and user's accuracy = 90% and 70% for the 1924 and 1986 fires, respectively). In 1924, fire extent from the atlas was greater than for fire scars, but the reverse was true for 1986. In 1986, fire extent interpreted from the delta normalized burn raton derived from pre-and post-fire satellite imagery was similar to that inferred from the fire-scar record (producer's accuracy = 92%, user's accuracy = 88%). In contrast, agreement between fire-scar and fire-atlas records was poor at the Washington sites. Fire atlasses are the most readily available source of information on the extent of late 20th century fires and the only source for the early 20th century. While fire atlases capture broad patterns useful at the regional scale, they should be field validated and used with caution at the local scale.