Sea duck (tribe mergini) populations have experienced significant declines in recent years, though underlying causes are poorly understood. Information on population demographic parameters may provide insight for wildlife managers seeking to maintain sustainable harvest. However, population monitoring capacity for sea ducks is limited relative to other waterfowl species because of their remote breeding distribution. The United States Fish and Wildlife Service organizes a Parts Collection Survey (PCS), which estimates recruitment in sea duck populations using age ratios (juveniles/adult), though estimates are biased because harvest vulnerability differs between age cohorts. We used a photo survey to calculate estimates of annual recruitment for long‐tailed duck (Clangula hyemalis), black scoter (Melanitta americana), surf scoter (M. perspicillata), and white‐winged scoter (M. deglandi; hereafter sea ducks) populations in eastern North America. Our team of surveyors collected photos of flighted sea ducks from shore and by boat in 11 states and 1 province from 15 October–15 December annually in 2019–2022. We classified photographed birds according to age and sex and calculated juvenile proportions of each species using a Bayesian binomial model. To compare photo survey estimates with PCS estimates, we used a paired t‐test organized by year. We found strong evidence that PCS estimates of juvenile proportions were greater than photo survey estimates for 3 sea duck species, indicating a consistent positive bias in PCS driven by harvest vulnerability. We also derived novel estimates of juvenile harvest vulnerability using the mean difference between within‐year estimates. Our work demonstrated the photo survey methodology produced annual recruitment estimates for 4 poorly monitored waterfowl populations with greater sample sizes and reduced systematic biases relative to existing methods; we recommend managers continue to adopt this approach in future years with additional consideration given for spatial representation and refinement of image classification procedures for long‐tailed duck estimates.