AbstractA central finding in decision neuroscience is that BOLD activity in several regions, including ventral striatum and ventromedial prefrontal cortex, is correlated with the subjective value of the option being considered, and that BOLD activity in these regions can predict choices out of sample, even at the population-level. Here we show, across two different decision making tasks in a large sample of subjects, that these BOLD value-correlates are intrinsically history dependent. If the subjective value of the previous offer was high, the signal on the current trial will be lower, and vice versa. This kind of history dependency is distinct from previously described adaptation or repetition suppression effects, but instead is of the form predicted by theories of efficient coding such as time-dependent cortical normalization. In terms of practical application, since value-based choice behavior does not exhibit the same history dependence, neural prediction studies may exhibit systematic errors without accounting for history effects. The data-driven, interpretable, whole-brain prediction approach we use to identify history effects also illustrates one way to adjust predictions for neural history dependency.