We have studied the effect of selective listening on the neuromagnetic evoked activity of the human auditory cortex. In the word categorization experiment the stimuli were 5-letter words, each beginning with (k). Half of them were targets, i.e., names of animals or plants, and half other meaningful Finnish words. In the duration discrimination experiment equiprobable tones of 425 ms (targets) or 600 ms duration were presented. In both experiments the interstimulus interval (ISI) was 2.3 s and the stimuli of the two classes were presented randomly. Subjects either ignored the stimuli (reading condition) or counted the number of targets (listening condition). The magnetic field over the head was measured with a 7-channel 1st-order SQUID-gradiometer. The stimuli evoked a transient response followed by a sustained field. The transient response did not differ between the two conditions but the sustained field was significantly larger in the listening than reading condition; the increase began 120-200 ms after stimulus onset and continued for several hundred milliseconds. The equivalent source locations of both transient and sustained responses agreed with activation of the supratemporal auditory cortex. In the dichotic listening experiment 25-ms square-wave stimuli were presented randomly and equiprobably either to the left or to the right ear at an ISI of 0.8-1 s, either alone or in presence of a speech masker. Counting the stimuli of either ear resulted in differences between responses to relevant and irrelevant sounds. The difference began 140-150 ms after stimulus onset and peaked at 200-240 ms. During monaural speech masking, N100m was larger for attended than ignored stimuli.(ABSTRACT TRUNCATED AT 250 WORDS)