Quantitative trait locus (QTL) mapping is frequently used in evolutionary studies to understand the genetic architecture of continuously varying traits. The majority of studies have been conducted in specially created crosses, in which genetic differences between parental lines are identified by linkage analysis. Detecting QTL segregating within populations is more problematic, especially in wild populations, because these populations typically have complicated and unbalanced multigenerational pedigrees. However, QTL mapping can still be conducted in such populations using a variance components mixed model approach, and the advent of appropriate statistical frameworks and better genotyping methods mean that the approach is gaining popularity. In this study it is shown that all studies described to date report evidence of QTL of major effect on trait variation, but that these findings are probably caused by inflated estimates of QTL effect sizes due to the Beavis effect. Using simulations I show that even the most powerful studies conducted to date are likely to give misleading descriptions of the genetic architecture of a trait. I show that an interpretation of a mapping study of beak color in the zebra finch (Taeniopygia guttata), that suggested genetic variation was determined by a small number of loci of large effect, which are possibly maintained by antagonistic pleiotropy, is likely to be incorrect. More generally, recommendations are made to how QTL mapping can be combined with other approaches to provide more accurate descriptions of a trait's genetic architecture.