Quantitative trait locus (QTL) studies in plants frequently employ phenotypic data on a population of lines (doubled haploid lines, recombinant inbred lines, etc.) tested in multiple environments. An important feature of such data is the genetic correlation among observations on the same genotype in different environments. Detection of QTL-by-environment interaction requires tests which take this correlation into account. In this article, a comparison was made of the properties of several such tests by means of simulation. The results indicate that a split-plot analysis of variance (ANOVA), being an approximate method, tends to be too liberal under departures from the Huynh-Feldt condition. A standard two-way ANOVA, which ignores genetic correlation, yields inappropriate tests and should be avoided. In contrast, mixed model approaches as well as univariate and multivariate repeated-measures ANOVA yield valid results. This supports the use of a flexible mixed model framework in more complex settings, which are difficult to tackle by repeated-measures ANOVA. .