A well-designed sampling scheme is critical for obtaining accurate results from population genetic studies. Larval samples contain only the genetic material of successful breeders, often of a single year, and may be biased towards particular families. To quantify the bias of using larval samples to infer population and landscape genetic structure and explore how this bias may be reduced using sibship analysis, we analysed eight microsatellite loci from 484 tissue samples of larvae and adults of Columbia spotted frogs (Rana luteiventris) and long-toed salamanders (Ambystoma macrodactylum) at nine breeding sites in north Idaho. Differences in allele frequencies between adult and larval samples were not detected after full-siblings were removed from the larval data set for Columbia spotted frogs; for long-toed salamanders, these differences remained at two out of four ponds. Data from Columbia spotted frog larvae indicated higher levels of differentiation among populations (median difference in FST = 0.020, P < 0.01), as predicted by population genetic theory, whereas data from larval samples of long-toed salamanders showed some evidence of lower levels of differentiation among populations (median difference in FST = 0.012, P = 0.06). For both species, removing all but one individual from each full-sibling family led to parameter estimates that were closer to those calculated from adult samples for both population and landscape genetic measures. Removal of full-siblings is likely to improve estimates of population genetic parameters; however, knowledge of the species’ breeding system is essential for understanding additional sources of bias when inferring population genetic structure from larval samples.