Estimating components of detection probability is crucial to improving the design of aerial surveys for wildlife populations, and this is especially true for species of marine mammals that are threatened or endangered. To evaluate the probability that Florida manatees (Trichechus manatus latirostris) will be detected by observers during aerial surveys, we conducted 6 series of survey flights, during mornings and afternoons on 14–16 consecutive days over the Tampa Electric Company's (TECO) Big Bend power plant discharge canal in Tampa Bay, Florida, USA (winter 2000 through 2003). Our objective was to understand how our ability to detect manatees at a winter aggregation site affects aerial survey counts, so that we may improve techniques for estimating manatee population size. We estimated the probability that manatees would be present at the warm-water discharge of the plant during winter cold fronts and estimated the overall detection probability of manatees present at the plant and the 2 components that make up the probability of detection (the probability of being available and the probability of being detected given they are available). We used telemetry tags and marker flags (n = 15) to facilitate capture–recapture analyses. The probability that marked manatees would be at the plant varied from 48% to 68% across flight series and was inversely related to the ambient water temperature. Based on sightings of marked animals, estimates of the overall probability of detecting a manatee ranged from 45% to 69% across flight series (x̄ = 58%, n = 6). The probability that a manatee would be available to an observer ranged from 73% to 94% across flight series (x̄ = 83%) but was constant among years (83%, 81%, and 78%; x̄ = 81%). The probability that an available manatee would be detected by an aerial observer was variable across flight series (55–95%) and years (73%, 86%, and 66%, x̄ = 73%). Independent estimates of the probability that a manatee would be available to the observer on one pass were obtained from time–depth data loggers and ranged from 5% to 33% (x̄ = 19%, SE = 3.7%), and the probability that a manatee would be available during ≥1 of 10 passes ranged from 41% to 98% (x̄ = 88%, 95% confidence bounds 0.71–0.95). We adjusted survey counts using measures of detectability. Although corrected counts presented here are site-specific, adjusting counts based on detection probability will greatly improve reliability of population estimates from all aerial surveys. Special sampling to estimate components of detection probability should be built into all aerial surveys to ensure that reliable and unbiased information on species abundance is used to evaluate wildlife populations.