Although hundreds of genetic association studies of ischaemic stroke have been published, the failure to replicate associations has led to scepticism about their findings. Possible explanations for this failure are: (1) a false-positive association in the initial study; (2) a false-negative association in a replication study; (3) methodological differences (e.g. study populations or study designs). We review underlying causes for replication failure, such as small sample size, multiple testing and publication bias, and methods to deal with these problems. We also make suggestions about the design of genetic association studies in ischaemic stroke with regard to stroke subtype classification, candidate pathways, subgroups, intermediate phenotypes and potential clinical impact.