Studies of trilled vocalizations provide a premiere illustration of how performance constraints shape the evolution of mating displays. In trill production, vocal tract mechanics impose a trade-off between syllable repetition rate and frequency bandwidth, with the trade-off most pronounced at higher values of both parameters. Available evidence suggests that trills that simultaneously maximize both traits are more threatening to males or more attractive to females, consistent with a history of sexual selection favoring high-performance trills. Here, we identify a sampling limitation that confounds the detection and description of performance trade-offs. We reassess 70 data sets (from 26 published studies) and show that sampling limitations afflict 63 of these to some degree. Traditional upper-bound regression, which does not control for sampling limitations, detects performance trade-offs in 33 data sets; yet when sampling limitations are controlled, performance trade-offs are detected in only 15. Sampling limitations therefore confound more than half of all performance trade-offs reported using the traditional method. An alternative method that circumvents this sampling limitation, which we explore here, is quantile regression. Our goal is not to question the presence of mechanical trade-offs on trill production but rather to reconsider how these trade-offs can be detected and characterized from acoustic data.