Underwater observation is a widely used fish-sampling method, but capture efficiencies of this method are often unknown. For accuracy, survey counts require correction by measuring capture efficiencies of sampling protocols. Capture efficiencies for underwater observation were calculated for three small imperiled fishes—Etheostoma sitikuense (Citico Darter), Noturus flavipinnis (Yellowfin Madtom), and Noturus baileyi (Smoky Madtom)—using modified mark-recapture methods. Fishes were tagged with visual implant elastomer tags, released at sites within Abrams Creek in the Great Smoky Mountains National Park, and then recaptured. Efficiencies were calculated by comparing numbers of released individuals to recaptures. In the propagation facility, tag retention was 100 percent, and no post-tagging mortality was observed. Capture efficiency (CE = 0.12) was low for all species and potentially influenced by predation upon marked fish, emigration of fish from sites, or difficulty in sampling some habitats. Thus, population sizes may be larger than observed due to low capture efficiencies. Our results highlight challenges to estimating capture efficiencies for imperiled fishes when using underwater observational methods.