Principles from particle-motion physics were applied to recurring problems of the interpretation of stratigraphic charcoal data: (1) fires within catchments of lakes often produce no record in fossil-charcoal curves and (2) periods characterized by no local fire (e.g., 20th-century fire suppression) often display as much charcoal as times when local fire was frequent. Quantitative theory on source area, transport, deposition, and sampling of charcoal shows the relationship between particle sizes counted by alternative methods of charcoal analysis (pollen slides for particles 5–80 μm in diameter, petrographic thin sections for particles 50–10,000 μm diameter) and charcoal diagrams. The relationship between diameter and critical and deposition velocities results in fundamental aerodynamic differences between the sizes of particles quantified by the two methods. Charcoal recorded on pollen slides is of a size that is difficult to lift, but once entrained it remains in suspension. Thin-section charcoal is lifted at relatively low wind velocities, but it is not suspended at normal surface wind speeds. Thermal buoyancy during fire lofts charcoal above the forest canopy, depending on particle size and wind speed. Pollen-slide charcoal sizes are underrepresented near the fire, because they remain in suspension and are preferentially exported from the burn area. Thin-section charcoal is convected to lower heights on average and is deposited nearby. Following fire, thin-section charcoal is redistributed locally by wind and thus may enter lakes. Because of cohesive forces and aerodynamics, more pollen-slide charcoal remains on the ground, and less enters lakes. Source areas for pollen-slide charcoal are subcontinental to global, and diagrams of pollen-slide charcoal are biased toward nonlocal charcoal. They can be used to interpret importance of fire for broad spatial and temporal scales. Thin-section charcoal represents, the catchment fire regime. Simulation models that generate charcoal during fire, mix sediments, and then sample at specified intervals indicate that (1) in the absence of sediment mixing the pollen-slide method should consistently resolve individual fires that occur with an expectation of >30–50 yr, (2) unless samples are continuous, neither method will produce useful estimates of fire frequency, and (3) even a modest amount of sediment mixing will obscure the signal.