Estimating demographic values and rates for populations of cryptic stream species frequently is difficult because of prohibitively low capture probabilities. When assessing cryptic populations, researchers often are forced to make simplifying assumptions that could alter their conclusions about a population. We constructed models based on different assumptions about cohort structure, behavioral responses to capture, temporary emigration, and survival in a cryptic, larval population of red salamanders (Pseudotriton ruber) and fitted models to capture-mark-recapture data. Overall, models based on 2 cohorts were favored over models based on 1 cohort. Models based on assumptions of constant survival, behavioral responses to capture, and random temporary emigration were ranked higher than models lacking these assumptions. Consistent behavioral responses to capture demonstrated that using uncorrected counts to assess trends for this, and perhaps other larval amphibian populations, yields misleading results. Counts that are not corrected for trap-shy behavior may inherently show negative temporal trends. Temporary emigration was a critical assumption when describing larval salamander demography because only 27% of the larvae were active on the surface (the rest were in substratum habitats). Our study demonstrates the importance of making appropriate assumptions about demographic parameters and shows how population models can quantify aspects of the natural history of cryptic species.