Land managers generally assume that postfire surveys conducted by fire personnel are effective at quantifying mortality or injury of box turtles (Terrapene spp.) because individuals should be readily observable in burned areas. Box turtle surveys conducted by humans in unburned habitats can be ineffective, yet little information exists on the efficacy of postfire surveys. We quantified detection probability of eastern box turtles (Terrapene carolina carolina) shortly after a prescribed fire in southwestern Michigan, USA, during May 2015. Immediately prior to a May (growing season) fire, we confirmed that 7 adult box turtles fitted with radiotransmitters occupied the proposed burn area. Two days after the burn we reconfirmed turtle locations and subsequently conducted 6 independent visual‐encounter surveys through 2 burned areas (0.75 and 1.0 ha) that contained telemetered turtles. For these 12 surveys, we found that average detection probability per survey was low (0.11, SE = 0.09) and highly variable among observers (range = 0.00–0.50). We found that individual turtles directly exposed to fire remained buried for up to 12 hr after the fire was extinguished and were thus unavailable for detection immediately after the burn. Further confounding postfire survey results, buried turtles rapidly moved to unburned areas after emerging from their subterranean refugia. Our results suggested that typical visual‐encounter surveys conducted for eastern box turtles after prescribed burning do not accurately reflect occupancy status or fire‐caused mortality.