Allozyme-associated heterosis has been repeatedly observed in marine bivalves, but its genetic origin remains debatable. A simple explanation is direct overdominance at the enzyme loci scored. An alternative is associative overdominance due to partial inbreeding, affecting the whole genome. The two hypotheses yield different predictions concerning (i) locus-specific effects, (ii) the relationship between heterozygosity and the variance in fitness, and (iii) the expected form of the relationship between the multilocus genotype and mean fitness. The relationship between heterozygosity and growth, a component of fitness, is here analysed in Spisula ovalis (1669 individuals, 9 loci), using statistical models designed to test these predictions. In contrast to most other bivalves, S. ovalis shells display clear annual growth lines allowing accurate quantification of individual age and growth. Our results show (i) that there is no evidence for locus-specific effects, (ii) that the variance in growth decreases significantly when heterozygosity increases, and (iii) that growth is better predicted by a genetic variable optimized for inbreeding than by a variable optimized for overdominance. In addition, the heterozygosity–growth relationship displays a significant variation among annual cohorts, being more pronounced in young cohorts. Although the need to pool alleles and the occurrence of null alleles may limit the efficiency of some of the models used (especially for result (iii)), our results suggest that the heterozygosity–growth relationship is due to inbreeding effects.