Background: As echocardiographic left ventricular mass does not scale linearly with body surface area, indexing of left ventricular mass to height raised to the power exponent b has been proposed (‘allometric indexing’). Pooling data from males and females has been reported to lead to a higher value of b than when each sex is considered separately. We wished to determine if this is due to within-sex clustering, as this would change both how optimal b should be determined and how indexing should be performed. Methods: Echocardiograms were performed in rigorously screened healthy subjects (96 men, 133 women), and b was calculated as the relationship between left ventricular mass and body height. Results: Clustering was detected within men and women through analysis of regression model residuals, violating assumptions of independence. A hierarchical model was built to allow for clustering, and compared with treating males and females as separate populations using Bayesian regression. Bayes factor analysis showed greatest credibility of the clustered model with optimal b determined as 1.64 (0.68–2.82) and 0.96 (0.17–1.73) in men and women, respectively. Conclusion: Data from men and women are best suited to multilevel modelling, and b should therefore not be estimated after pooling. This indicates that allometric indexing may require sex-specific values either for the power exponent or for any subsequently derived reference ranges.