The emergence of high-throughput, genome-scale approaches for identifying and genotyping DNA variants has been a catalyst for the development of increasingly sophisticated whole-genome association and genomic prediction approaches, which together have revolutionized the study of complex traits in human, animal, and plant populations. These approaches have uncovered a broad spectrum of genetic complexity across traits and organisms, from a small number of detectable loci to an unknown number of undetectable loci. The heritable variation observed in a population is often partly caused by the segregation of one or more large-effect (statistically detectable) loci. Our study focused on the accurate estimation of the proportion of the genetic variance explained by such loci (p), a parameter estimated to quantify and predict the importance of causative loci or markers in linkage disequilibrium with causative loci. Here, we show that marker-associated genetic variances are systematically overestimated by standard statistical methods. The upward bias is purely mathematical in nature, unrelated to selection bias, and caused by the inequality between the genetic variance among progeny and sums of partitioned marker-associated genetic variances. We discovered a straightforward mathematical correction factor (k M ) that depends only on degrees of freedom and the number of entries, is constant for a given experiment design, expands to higher-order genetic models in a predictable pattern, and yields bias-corrected estimates of marker-associated genetic variance and heritability.