1. Measures of body mass and length are commonly used to derive indices of condition, which are often assumed to reflect the energetic state of an animal and, by extension, to predict its fitness. However, the relationship between condition and fitness is rarely quantified, and the appropriate method(s) for deriving indices of condition are frequently debated. 2. Data from a decade of mark-recapture studies involving over 4,600 individual lizards (Anolis sagrei) and 41 replicates of selection across seven populations were used to test the common assumption that condition, as estimated from body mass and length, is a strong predictor of adult survival, an important component of fitness. Inferences about natural selection were compared between two alternative indices of condition: the popular residual index (Ri), and the scaled mass index (Mi), which was recently proposed as a more appropriate method. 3. Linear, quadratic, and correlational selection gradients obtained using Ri and Mi were highly correlated with one another. Relative to variance in selection among replicates and to error associated with the estimation of selection gradients, variance due to the use of alternative condition indices was minor and effectively negligible. 4. Contrary to the intuitive prediction that individuals in better condition should exhibit higher survival, there was no evidence for strong or consistent linear (directional) selection for higher condition indices in either males or females. Significant quadratic (stabilizing or disruptive) selection on condition was similarly rare in both sexes. Correlational selection favored combinations of large size and high condition in males, but not females. 5. Collectively, these results indicate that inferences about natural selection may be robust to the choice between indices of condition, but that indices of condition can be unreliable as proxies for fitness, particularly when relationships between fitness components and condition are contingent upon interactions with other traits.