© 2015 by the American Psychosomatic Society. Research on eating relies on various indices (e.g., stable, momentary, neural) to accurately reflect food-related reactivity (e.g., disinhibition) and regulation (e.g., restraint) outside the laboratory. The degree to which they differentially predict real-world consumption remains unclear. Further, the predictive validity of these indices might vary depending on whether an individual is actively restricting intake. Methods We assessed food craving reactivity and regulation in 46 healthy participants (30 women, 18-30 years) using standard measurements in three modalities: a) self-reported (stable) traits using surveys popular in the eating literature, and b) momentary craving ratings and c) neural activation using aggregated functional magnetic resonance imaging data gathered during a food reactivity-And-regulation task. We then used these data to predict variance in real-world consumption of craved energy-dense "target" foods across 2 weeks among normal-weight participants randomly assigned to restrict or monitor target food intake. Results The predictive validity of four indices varied significantly by restriction. When participants were not restricting intake, momentary (B = 0.21, standard error [SE] = 0.05) and neural (B = 0.08, SE = 0.04) reactivity positively predicted consumption, and stable (B =-0.22, SE = 0.05) and momentary (B =-0.24, SE = 0.05) regulation negatively predicted consumption. When restricting, stable (B = 0.36, SE = 0.12) and neural (B = 0.51, SE = 0.12) regulation positively predicted consumption. Conclusions Commonly-used indices of regulation and reactivity differentially relate to an ecologically-valid eating measurement, depending on the presence of restriction goals, and thus have strong implications for predicting real-world behaviors.