One drawback of functional magnetic resonance imaging (fMRI) is that the subject must endure intense noise during testing. We examined the possible role of such noise on the activation of early visual cortex during visual mental imagery. We postulated that noise may require subjects to work harder to pay attention to the task, which in turn could alter the activation pattern found in a silent environment. To test this hypothesis, we used positron emission tomography (PET) to monitor regional Cerebral Blood Flow (rCBF) of six subjects while they performed an imagery task either in a silent environment or in an "fMRI-like" noisy environment. Both noisy and silent imagery conditions, as compared to their respective baselines, resulted in activation of a bilateral fronto-parietal network (related to spatial processing), a bilateral inferior temporal area (related to shape processing), and deactivation of anterior calcarine cortex. Among the visual areas, rCBF increased in the most posterior part of the calcarine cortex, but at level just below the statistical threshold. However, blood flow values in the calcarine cortex during the silent imagery condition (but not the noisy imagery condition) were strongly negatively correlated with accuracy; the more challenging subjects found the task, the more strongly the calcarine cortex was activated. The subjects made more errors in the noisy condition than in the silent condition, and a direct comparison of the two conditions revealed that noise resulted in an increase in rCBF in the anterior cingulate cortex (involved in performance monitoring) and in the Wernicke's area (required to encode the verbal cues used in the task). These results thus demonstrate a nonadditive effect of fMRI gradient noise, resulting in a slight but significant effect on both performance and the neural activation pattern.