Heterogeneity in fitness components consists of fixed heterogeneity due to latent differences fixed throughout life (e.g. genetic variation), and dynamic heterogeneity generated by stochastic variation. Their relative magnitude is crucial for evolutionary processes, as only the former may allow for adaptation. However, the importance of fixed heterogeneity in small populations has recently been questioned. Using neutral simulations (NS), several studies failed to detect fixed heterogeneity, thus challenging previous results from mixed models (MM). To understand the causes of this discrepancy, we estimate the statistical power and false positive rate of both methods, and apply them to empirical data from a wild rodent population. While MM show high false positive rates if confounding factors are not accounted for, they have high statistical power to detect real fixed heterogeneity. In contrast, NS are also subject to high false positive rates, but have always low power. Indeed, MM analyses of the rodent population data show significant fixed heterogeneity in reproductive success, whereas NS analyses do not. We suggest that fixed heterogeneity may be more common than is suggested by NS, and that NS are useful only if more powerful methods are not applicable and if they are complemented by a power analysis.