In line with inbreeding theory, genetic diversity at a set of molecular markers may explain variation in fitness-associated traits in partially inbred populations, and such associations will appear as ‘genotype–fitness correlations’. An individual genetic diversity index specifically used for microsatellites is ‘mean d2’, i.e. the mean squared distance between alleles. The original hypothesis for mean d2–fitness correlations assumes that mean d2 captures fitness effects at both ends of the inbreeding–outbreeding spectrum. This hypothesis received strong criticism from work showing that even a plain diversity estimate such as multi-locus heterozygosity (MLH) outperforms mean d2 as a predictor of the inbreeding coefficient and fitness in most realistic situations. Despite this critique, the mean d2-approach is still used frequently in ecological and evolutionary research, producing results suggesting that mean d2 sometimes provides a stronger prediction of fitness than does MLH. In light of the critique, such results are unexpected, but potential explanations for them may exist (at least hypothetically), including scenarios based on close linkage and recent admixture. Nevertheless, a major caveat is that it is very difficult to predict a priori if mean d2 will improve the genotype–fitness correlation, which in turn makes objective interpretations difficult. Mean d2–fitness associations are potentially interesting, but the fact that we cannot easily understand them is problematic and should be thoroughly addressed in each study. Therefore, instead of hastily reached interpretations of mean d2–fitness correlations, conclusions need support from complementary analyses, e.g. verifying admixture of genetically structured populations.