Mental imagery is an important tool in the cognitive control of emotion. The present study tests the prediction that visual imagery can generate and regulate differential fear conditioning via the activation and prioritization of stimulus representations in early visual cortices. We combined differential fear conditioning with manipulations of viewing and imagining basic visual stimuli in humans. We discovered that mental imagery of a fear-conditioned stimulus compared to imagery of a safe conditioned stimulus generated a significantly greater conditioned response as measured by self-reported fear, the skin conductance response, and right anterior insula activity (experiment 1). Moreover, mental imagery effectively down- and up-regulated the fear conditioned responses (experiment 2). Multivariate classification using the functional magnetic resonance imaging data from retinotopically defined early visual regions revealed significant decoding of the imagined stimuli in V2 and V3 (experiment 1) but significantly reduced decoding in these regions during imagery-based regulation (experiment 2). Together, the present findings indicate that mental imagery can generate and regulate a differential fear conditioned response via mechanisms of the depictive theory of imagery and the biased-competition theory of attention. These findings also highlight the potential importance of mental imagery in the manifestation and treatment of psychological illnesses.

While modern theories of emotion emphasize the role of higher-order cognitive processes such as semantics in human emotion, much research into emotional learning has ignored the potential contributions of such processes. This study aimed to determine whether emotional learning affects semantic representations of words independent of perceptual features by assessing whether fear conditioning to a neutral word generalises across languages in bilingual participants. Two sessions differing according to the reinforced language were performed by English-Spanish bilinguals. In each session, a neutral word was reinforced by an electrical shock whereas its equivalent in the other language was never paired with shock. Across two sessions within our sample, we found replicable evidence that fear conditioning consistently transferred to the non-reinforced language as measured by both self-reported fear and electrodermal activity, irrespective of the conditioned language. Our findings extend knowledge about the role of semantic similarity in fear generalisation and highlight the importance of higher-order cognitive processes in human emotions.