Selection of prey that are small and in poor body condition is a widespread phenomenon in terrestrial predator-prey systems and may benefit prey populations by removing substandard individuals. Similar selection is widely assumed to operate in aquatic systems. Indeed, size-selective predation is a longstanding and central tenet of aquatic food web theory. However, it is not known if aquatic predators select prey based on their condition or state, compared to their size. Surprisingly, no comparable information is available for marine systems because it is exceedingly difficult to make direct observations in this realm. Thus the role of body condition in regulating susceptibility to predation remains a black box in the marine environment. Here we have exploited an ideal model system to evaluate selective predation on pelagic marine fish: comparing characteristics (fork length, mass corrected for fork length) of fresh, whole, intact juvenile Pacific salmon delivered by a seabird to its single nestling with salmon collected concurrently in coastal trawl surveys. Three species of juvenile salmon (Oncorhynchus spp.) are consumed by provisioning Rhinoceros Auklets (Cerorhinca nionocerata); an abundant, colonial, pursuit-diving seabird. Samples were collected from multiple colonies and fisheries surveys in coastal British Columbia in two years. As predicted, Auklets preyed on small individuals in poor condition and consistently selected them at levels higher than their relative availability. This is the first study to provide direct evidence for both size- and condition-selective predation on marine fish in the wild. We anticipate that our results will be a starting point in evaluating how selective predation may structure or influence marine fish populations and bridges a fundamental incongruity between ecological theory and application; although "bigger is better" is considered a fundamental tenet of marine food webs, marine predators are often assumed to consume indiscriminately.