Résumé : 1. Cerebral potentials evoked by random sequences of electrical stimuli to four fingers were recorded in intact man performing selective attention tasks. Eye movements and other artifacts were excluded from the averaged traces. Different finger stimuli were designated as targets to be mentally counted in alternate runs of each experiment. The high mean random rate of stimuli (150/min) fully involved the processing capacities of the subject. Vigilance changes or differential expectancy effects were excluded by the reciprocal random design with four different sensory channels. Task-related enhancements of somatosensory evoked potentials (s.e.p.) components were estimated by comparison with the s.e.p.s to physically identical finger stimuli recorded in runs when the subject attended signals in the opposite hand. The experimental design avoided subject's fatigue.2. The primary s.e.p. components N(20) and P(45) were not significantly influenced and this excluded centrifugal gating of the corticipetal signals as a mechanism.3. The earliest task-related changes in s.e.p. occurred 55-135 msec (mean 77.7 msec) after the target finger stimuli. In most cases the negative N(140) component was markedly enhanced both for target signals and for non-targets in the adjacent finger of the same hand. However, in several subjects the targets elicited a positive P(100) component instead. Both N(104) and P(100) were larger at the contralateral parietal focus than ipsilaterally. They were definitely smaller at the vertex and frontal scalp locations.4. Enhancements of N(140) were not observed in similar random four-finger experiments carried out at a 4 times slower mean rate, but they occurred in a bisensory paradigm with finger shocks and acoustic clicks at that slower rate.5. A large positive P(400) component was only elicited by target stimuli. Its voltage was maximum over the parietal region and was equal on both sides.6. At least three categories of components can be differentiated in the cortical s.e.p. on the basis of their time domains (roughly 18-70 msec, 70 to 200-250 msec and over 200 msec after the finger stimuli), cerebral hemispheres topography and cognitive parameters. Verbal instructions defining specific perceptual tasks can to a large extent switch on and off the components of the second and third categories when the processing resources of motivated subjects are fully committed in a well designed forced paced paradigm. In certain individuals physiological evidence for a different ;stimulus set' processing of target (P(100)) and non-target (N(140)) signals was documented for the first time.