Ecological research using biometric data is only sound if biometrics themselves are accurate and not confounded by measurement error. Given concerns about the accuracy of biometrics taken directly (physical measurement of animals), digital measurement of photographs is often advocated, particularly for small or live specimens. However, there is currently limited understanding of intra- and inter-observer variability of such measurements (i.e. variability of multiple measurements of the same specimen by the same observer, and variability of multiple measurements of the same specimen by different observers, respectively). We took biometrics (two linear, two curvilinear, two angular) from moth photographs using standard image software and calculated two fluctuating asymmetry measures. Inter-observer variability was always higher than intra-observer variability. Measurement error was low for linear/curvilinear measurements (< ∼4%), but high for angular variables (52%) and asymmetry measures (45%). Measurement precision correlated positively with trait size. Variability caused significant differences in mean measurements inter-specifically for half the biometrics; there were no significant intra-specific differences. We discuss the implications of our findings for research using photographically-derived biometrics and offer recommendations for reducing measurement error.