Despite decades of research, there is still little known about the natural abundances and ecological importance of stream salamander larvae in eastern North America. Widely used methods for sampling larvae and estimating population densities can be significantly biased, which may have implications for monitoring efforts and studies addressing the effects of salamanders on stream processes such as nutrient dynamics. We compared the efficacy of two methods of capture, passive leaf litter trapping and dip netting, and performed mark–recapture (M-R) on larvae occurring in three adjacent streams. There was a significant size bias associated with at least one of our capture methods. Leaf litter traps captured a higher proportion of larger individuals (>15 mm SVL), whereas dip netting yielded a greater proportion of the smaller size class (≤15 mm). Density estimates were 1–2 orders of magnitude greater than estimates provided by other studies; however, count indices were comparable to those previously reported. Density estimates from M-R ranged from 23–169 salamander larvae per m², which was remarkably consistent with estimates of 75–137 larvae per m² from a contemporaneous study using nearly identical methodology in Appalachian streams in North Carolina. The coefficient of variation for mean density estimates generated from M-R was greater than that detected with count data, indicating that the use of replicated counts without adjustment for detection may be less effective for measuring spatial and temporal variation in larval stream salamander numbers.