There is increasing interest in the basis of commonly observed heterozygosity–fitness correlations (HFCs). Two models appear possible, a genome-wide effect due to inbreeding depression, and a single-locus effect due to chance linkage to a gene(s) experiencing balancing selection. Recent studies suggest that the latter tends to be more important in the majority of studies, but tests for the presence of single-locus effects tend to be rather weak. One of the problems is that the linkage disequilibrium between a microsatellite and a nearby gene experiencing balancing selection is never likely to be 100%. With this in mind, we conduct stochastic simulations aimed at determining the conditions under which single-locus HFCs may develop. We also suggest a new approach that could offer improved detection of HFCs but which also offers a more general method for detecting genotype–fitness correlations. Our method is based on looking for the maximum possible strength of association between genotype and fitness, and then asking whether randomized data sets are able to generate similarly strong associations. This method is tested on both simulated and real data. In both cases, our method generates greater levels of significance than current tests. Applied to previously published data from wild boar affected by tuberculosis, the method uncovers a strong single-allele association that is strongly predictive of whether the disease is localized or spreads throughout the body. We further suggest a simple method for dealing with the problem of population structure, and believe this approach will help to identify genomic regions associated with fitness.