It has routinely been assumed that cognitive effort carries a subjective cost. This assumption is both intuitive and explanatorily convenient, yet the relevant evidence is predominantly indirect, and little is known about neural processes responsible. The present research pursued a hypothesis that intrinsic cognitive costs arise from the engagement of executive control functions. An initial series of behavioral experiments confirmed that decision makers avoided courses of action with high control demands, including demands for working memory and task set reconfiguration. Neuroimaging experiments then examined neural correlates of the evaluation of cognitive costs. These studies were guided by an existing theory of prefrontal function, which identifies distinct computational roles for lateral and dorsomedial prefrontal cortex (LPFC and DMPFC). In the first fMRI experiment, cognitive costs were assessed at regular intervals via a subjective rating method while participants performed a sustained set-switching task. Activity in numerous brain regions fluctuated with performance errors and behavioral reaction times, but only in LPFC did brain activity predict ratings independently of these factors. A second, complementary fMRI experiment focused on individual differences in demand avoidance. The same LPFC region showed greater task-related activity in high-avoidance individuals, again implicating it in the registration of cognitive costs. Next, reanalyses were conducted of a previously published experiment, which had shown that cognitive costs caused neural reward responses to be discounted. On the basis of functional connectivity, the reanalysis identified LPFC as a potential source of the discounting effect. A final fMRI study focused directly on DMPFC. Participants made perceptual categorizations based on ambiguous evidence. Ambiguity level influenced activity in both DMPFC and striatum, and these neural responses directly predicted participants' avoidance of ambiguous trials. Overall, evidence suggests that cognitive costs involve both DMPFC and LPFC, and are related both to bottom-up (performance monitoring) and top-down (control engagement) signals. This research program explores how subjective disutility might emerge from cognitive information processing. Normatively, cognitive costs may guide agents toward efficient strategies, but may also impose soft limits on performance. This work advances our mechanistic understanding of this seemingly ubiquitous influence on behavior.