Auditory evoked potentials were recorded while participants attended to visually presented digits. The difficulty of the visual task was manipulated by requiring participants to process only the current digit (0-back) or both the current and the preceding digit (1-back). Tones deviating in frequency from standard tones elicited a frontal mismatch negativity peaking around 200 ms which did not vary with visual task. However, decreasing the visual task load enhanced a right-temporal positive wave peaking around 200 ms when tones were presented slowly, and a frontocentral negative wave peaking around 450 ms when tones were presented more rapidly. The degree to which task-irrelevant sounds are processed therefore depends on the degree to which a visual task engages attentional resources.