Snorkel surveys are frequently used to monitor stream-dwelling fish. Inferring local abundance from snorkel surveys is complicated by two primary factors: variable fish detection probabilities and the relative abundance of fish in habitat types below the recommended minimum depth for snorkeling. We examine these factors across three salmonid species (Oncorhynchus spp.), 4 years, and 113 location-years in Oregon coastal streams. We calibrate snorkel counts to mark-recapture estimates and develop mathematically explicit expressions that convert a new snorkel count into a probability density of abundance for streams containing mixed habitat types that were and were not snorkeled. Snorkelers detected 63%, 47%, and 39% of juvenile coho salmon (O. kitsutch), steelhead trout (O. mykiss), and cutthroat trout (O. clarkii clarkia) estimated by mark-recapture, respectively, but uncertainty within and among sampling units was high. Further, analytics developed here can be used to infer abundance and uncertainty for habitat types that were not snorkeled. Our quantification of uncertainty arising from using snorkel counts as a proxy for abundance will help managers balance biological risks with available resources.