Context Significant opportunities have been identified in the northern Australian beef industry that can improve efficiency and profitability by using composite or crossbred cattle and genomic selection. The improved performance of composite cattle is partly due to heterosis. One of the major genetic bases of heterosis is dominance. Traditionally, dominance is ignored in genetic evaluation but could improve the accuracy of breeding values and help maintain genetic diversity. Aims The aim of this study is to describe the impact of including a dominance relationship matrix with different parameterisation methods and including heterozygosity fraction on estimated breeding values for 400-day weight in a composite population. Methods Genotype and phenotype data were obtained from 2364 tropical composite animals and were imputed to 27 648 single nucleotide polymorphisms. Genetic parameters and breeding values were estimated for 400-day weight from a linear mixed model using a genomic relationship matrix, heterozygosity fraction and three different parameterisation methods for the dominance relationship matrix, including genotypic, classical and the natural and orthogonal interaction approach. Genetic parameters and breeding values where compared over the three different parameterisation methods. Key results The heritability for all models when heterozygosity was not fitted ranged from 0.25 to 0.35, with the genotypic dominance model having the lowest additive heritability. Including heterozygosity fraction in the model as a fixed covariate resulted in substantial (39–49%) reductions in dominance variance across all models but a minimal change in the additive variance and, therefore, heritability (0.29–0.35). Conclusions and Implications In a composite population, including heterozygosity fraction in the model was important due to directional dominance. When heterozygosity fraction was not included, the genetic variance was incorrectly partitioned, and the dominance estimates were biased. Including the dominance relationship matrix improved the accuracy of breeding values. Parameterisation methods for forming the dominance relationship matrix are largely a matter of what estimates are required from the models and convenience. The additive values were largely independent of dominance parameterisation when heterozygosity was in the model.