Replicable populations, such as panels of recombinant inbred or doubled haploid lines, are convenient resources for the mapping of QTL. To increase mapping power, replications are often collected within each RI line and a common way to analyze such data is to include in the QTL model only a single measurement from each line that represents the average among the replicates (a line means model). An obvious, but seldom explored, alternative, is to include every replicate in the model (a full data model). Here, we use simulations to compare these two approaches. Further, we propose an extension of the standard permutation procedure that is required to correctly control the type I error in mapping populations with nested structure.