In the last decade, research efforts in neuroimaging analysis, machine-learning algorithms, and formal reverse inference of mental states have led to great advances in decoding cognitive states from neuroimaging data. The advent of brain-decoding in psychological science poses the very exciting prospect of being able to measure latent processes such as value computation, belief representation, and decision-policy updates. However, a key impediment in the application of brain-decoders to psychology is their lack of interpretability. The focus of machine-learning algorithms on prediction rather than hypothesis testing has led to brain decoders that can predict behaviors well, but that may not clearly reveal what cognitive processes are being measured. Here, I aim to break this stalemate by incorporating cognitive theories to guide construction of neural predictors that provide not just predictive, but more valid measures of cognitive states in decision-making.