Determining how the frequency, severity, and extent of forest fires are changing in response to changes in management and climate is a key concern in many regions where fire is an important natural disturbance. In the USA the only national-scale fire severity classification uses satellite image change-detection to produce maps for large (>400 ha) fires, and is generated by the Monitoring Trends in Burn Severity (MTBS) program. It is not clear how much forested area burns in smaller fires or whether ground-based fire severity estimates from a statistical sample of all forest lands might provide additional, useful information. We developed a tree mortality based fire severity classification using remeasured tree data from 10,008 plots in a probabilistic survey of National Forests System (NFS) lands in Oregon and Washington, using 8 tree mortality and abundance metrics. We estimate that 12.5% (±0.7% SE) of NFS forest lands in the region experienced a fire event during 1993–2007, with an annual rate of 0.96% (±0.05%). An estimated 6.5% of forest lands burned at High Severity or Moderate Severity; 2.1% burned at Very Low severity or only experienced surface or understory fire. A total of 358 of the 507 burned plots were within the MTBS perimeters, with ∼45% having equivalent severity classifications; but for ∼51% of the plots the MTBS classifications suggested lower severity than the tree-mortality based classes. Based on events recorded on plots and the inventory design, we estimate that 20.9% of the forested NFS lands experiencing fires, either wildfires or prescribed burns, were not in the MTBS maps. Tree mortality based fire severity classifications, combined with remotely-sensed and management information on timing and treatments, could be readily applied to nationally-consistent Forest Inventory and Analysis (FIA) data to provide improved monitoring of fire effects anywhere in the USA sampled by remeasured FIA inventories.