The ability of populations to respond to natural or sexual selection, termed “evolvability” in quantitative genetics (Houle 1992; Lynch and Walsh 1998; Hansen 2006; Sniegowski and Murphy 2006; Hansen and Houle 2008; Pigliucci 2008), is contingent on the level of additive genetic variation underlying trait expression. Consequently, a common practice in quantitative genetics studies is to derive standardized measures of evolvability that allow comparisons among traits and taxa. In a landmark paper, Houle (1992) proposed that a dimensionless statistic (see also Charlesworth 1987), termed the coefficient of additive genetic variation (*CV*_{A}), is appropriate for such purposes. *CV*_{A} is simply

that is, the square root of the additive genetic variance (*V*_{A}) divided by the phenotypic mean of the trait (note that Houle (1992) expressed this quantity using a 100 multiplier). Unlike heritability, *CV*_{A} is a measure of additive genetic variation that is standardized by the trait mean and therefore independent of other sources of variance. It is precisely these properties that make *CV*_{A} suitable for comparative purposes.

Houle (1992) addressed a long-standing difficulty with the interpretation of patterns of genetic variation in fitness traits. Traits closely related to fitness, such as survival or fecundity, typically exhibit lower narrow-sense heritabilities (i.e., the ratio of additive genetic variance to total phenotypic variance) than traits under weak or stabilizing selection, such as morphological traits (Gustafsson 1986; Charlesworth 1987; Mousseau and Roff 1987; Roff and Mousseau 1987; Houle 1992; Falconer and Mackay 1996; Kruuk et al. 2000; Merila and Sheldon 2000). This pattern was traditionally interpreted as resulting from the depletion of genetic variation due to strong directional selection (Fisher 1930). By contrast, Houle (1992) showed that traits closely associated with fitness generally exhibit higher *CV*_{A}s, and thus higher, not lower additive genetic variability than those under weaker selection. Houle's (1992) data supported the view that traits closely associated with fitness have higher levels of residual variation (e.g., nonadditive genetic, maternal, and environmental variation, including error variation), thereby explaining their low heritabilities (Barton and Turelli 1989; Price and Schluter 1991).

Since the publication of Houle's (1992) paper, there has been a proliferation of studies reporting either mean-scaled additive genetic variances (predominantly *CV*_{A} but see below) or using these evolvability measures for comparisons, and this work has improved our understanding of the factors that contribute toward the maintenance of genetic variation. This work has led to the general consensus that traits tightly linked to fitness exhibit high levels of both genetic and residual variation (e.g., Merila and Sheldon 1999, 2000; Kruuk et al. 2000; McCleery et al. 2004; Coltman et al. 2005; Hansen et al. 2011). Furthermore, the realization that some traits harbor considerable additive genetic variance despite strong directional selection has fuelled the development of new theory, such as the genic-capture model used to address the lek paradox (Kotiaho et al. 2008), and more generally the maintenance of genetic variation in fitness traits (Rowe and Houle 1996; Tomkins et al. 2004).

Given the utility of *CV*_{A} for comparative studies of genetic variation, it is crucial that primary studies employ correct and consistent methods to estimate this parameter (or suitable alternatives; see below and Discussion). If mistakes are frequent, incorrectly calculated *CV*_{A}s are likely to have been reported in reviews or studies that compile or compare these values, thus potentially biasing and/or confounding the conclusions drawn from such studies. In an attempt to determine the extent to which mistakes in the calculation of *CV*_{A} occur in the literature, and their potential consequences, we have reviewed recent quantitative genetic studies that have reported this statistic. We also review important issues in relation to the use and limitations of *CV*_{A}.

In his original paper, Houle (1992) also described another mean-standardized additive genetic variance, termed *I*_{A}, as a measure of evolvability (Houle 1992; Hansen et al. 2011). *I*_{A} equals *CV*_{A}^{2} if *CV*_{A} is expressed as in equation (1)

Although *CV*_{A} and *I*_{A} are related, they are distinct quantities (Houle 1992), and a key advantage of *I*_{A} is that its numerical value can be interpreted as the expected proportional change under a unit strength of selection (see Hansen et al. 2003; Hereford et al. 2004; Hansen et al. 2011). For this reason, Hansen et al. (2011) recommend the use of *I*_{A} as a measure of evolvability. It is therefore likely that future research will shift focus from *CV*_{A} to *I*_{A}. Our review, however, focuses on *CV*_{A} because until now this coefficient has been used predominantly to report and compare evolvabilities. Nevertheless, given the relationship between *CV*_{A} and *I*_{A} and the fact that both involve mean scaling, both measures suffer from similar limitations and are prone to similar calculation errors. Therefore, our results and guidelines can be extended to both measures of evolvability.