Brood:pair ratios could provide an economical method for assessing spatial or temporal variation in waterfowl productivity, but such estimators are severely biased by incomplete detection of broods. We conducted 3 sequential counts of 1,357 waterfowl broods in northeastern North Dakota, USA, and used closed-population mark–recapture models to estimate total brood abundance while controlling for variation in detection probabilities (p). Blue-winged teal (Anas discors) broods had the lowest average detection probability (p = 0.305), whereas diving-duck broods had the highest average detectability (p = 0.571). Detection was generally highest in morning or evening, but temporal patterns varied among species and there was no survey window that maximized detection probabilities for all species. Detection probabilities averaged 0.108 (SD = 0.056) higher for an experienced observer versus an inexperienced observer. Detection probabilities were 0.044 higher for roadside versus walk-up surveys and increased with increasing brood size, total brood abundance, survey date, wind speed, temperature, cloud cover, and amount of time spent surveying each wetland. Detection probabilities declined with increasing wetland size and amount of tall peripheral vegetation. Our mark–recapture results indicated that a traditional unreplicated brood survey would have missed 67.5% of estimated broods, summed over all species. Use of closed-population mark–recapture techniques provided an effective method for reducing this bias and identifying and quantifying factors that reduce detection probabilities of waterfowl broods. We recommend that future brood surveys incorporate 2 or 3 temporally segregated replicate counts to allow for formal estimation of detection probabilities.