QTL mapping of plant trait genes using marker data from crosses between inbred lines has been very successful but fine-scale mapping is going to require an alternative approach. Association mapping depends on the linkage disequilibrium between marker and trait genes remaining after very many generations, rather than the few generations within F2 populations, backcrosses or even recombinant inbred lines. The success of association mapping, however, will depend on the availability of very dense marker sets, with SNPs appearing to be the best candidates. Even within the broad category of association mapping, plant geneticists need to decide whether to use random samples or parent-offspring pairs or trios. They also need to decide whether to test for trait association with allelic or genotypic data, with likely preference for genotypes for inbred species. There is also the need to decide between whole-genome scans or candidate gene searches or a two-stage design that incorporates some features of both. These various statistical issues, including the vexing one of false positive results with multiple testing, will be described in this talk and illustrated by reference to current activity in human genetics.