The effect of selective attention on activity of the right human auditory cortex was studied with a 24-channel planar SQUID-gradiometer. Two conditions were used, favoring either a late attention effect following N100m, or an early effect, overlapping with N100m. In experiment 1 (15 subjects), a randomized tone sequence of 1 and 3 kHz tones was delivered to the left ear with a constant interstimulus interval (ISI) of 405 msec. The subjects' task was to count infrequent longer tones of one of these pitches among shorter standards. An attention effect, called magnetic difference (Md), was found when the responses to the irrelevant standards were subtracted from those to the relevant standards. Md peaked at about 220 msec for the 1 kHz tones and at 195 msec for the 3 kHz tones. The equivalent source of Md was in the supratemporal auditory cortex, about 1 cm anterior to the source of N100m, and in the same location as the source of P200m. In experiment 2 (8 subjects) the paradigm was similar, except that the 1 kHz and 3 kHz tones were led to different ears with a random ISI of 240-300 msec. In this case Md started already at 30-40 msec, adding to the N100m deflection, and the sources of N100m and Md overlapped. Present results show that attention can modify the activity of two different areas in the supratemporal auditory cortex. We interpret both attention effects as alterations of the exogenous evoked response components: the earlier effect as changed activity in neurons underlying N100m to relevant tones and the later effect as a modification of P200m to irrelevant tones.