We used removal sampling (RS) and neutral red dye capture–mark–recapture (CMR) methods to estimate capture probabilities for larval Wood Frogs (Lithobates sylvaticus), Spring Peepers (Pseudacris crucifer), and first-year Green Frogs (Lithobates clamitans) in ponds in southern New Brunswick, Canada. We modeled capture probability as a function of environmental variables and tested whether marks were retained between surveys. We also performed simulations to understand the effect of survey effort, capture probability, and abundance on detection probabilities and the number of surveys needed to be confident of absence. Capture probabilities (P ± SD) were low and variable: Wood Frogs, P = 0.262 ± 0.128; Spring Peepers, P = 0.323 ± 0.241; and first-year Green Frogs, P = 0.159 ± 0.106. With the use of AICc, we determined a model with the proportion of non-Typha emergent vegetation to have a positive effect, whereas pond depth had a negative effort on capture probabilities in first-year Green Frog larvae. No covariate models were better than an intercept-only model for Wood Frogs or Spring Peepers. Observers missed the dye mark on 5–24% of the marked larvae. Simulations showed that at observed capture probabilities and low abundances (greater than 100 larvae) all species would be detected on average with a single survey of at least 20% of the pond, but more survey effort/repeat surveys would be required to detect smaller populations. We recommend that capture probabilities be estimated whenever abundance estimates are required.