The cognitive neuroscience of language relies largely on controlled experiments that are different from the everyday situations in which we use language. This review describes an approach that studies specific aspects of sentence comprehension in the brain using data collected while participants perform an everyday task, such as listening to a story. The approach uses ‘neuro‐computational’ models that are based on linguistic and psycholinguistic theories. These models quantify how a specific computation, such as identifying a syntactic constituent, might be carried out by a neural circuit word‐by‐word. Model predictions are tested for their statistical fit with measured brain data. The paper discusses three applications of this approach: (i) to probe the location and timing of linguistic processing in the brain without requiring unnatural tasks and stimuli, (ii) to test theoretical hypotheses by comparing the fits of different models to naturalistic data, and (iii) to study neural mechanisms for language processing in populations that are poorly served by traditional methods.