Human evoked-potential research on the neurophysiological substrate of selective attention is reviewed. Most of these studies report enhanced amplitudes of potentials evoked by attended (task-relevant, meaningful, important, etc.) stimuli the results of which are generally regarded as providing an electrophysiological correlate for selective attention. In accepting such claims, there appears to be two major procedural problems generally not satisfactorily solved in these studies: (1) the inability to reliably separate the specific and non-specific physiological changes concomitant with selective attention from each other; and (2) inadequacy of peripheral sensory control possibly inducing contaminating changes already at the level of the proximal stimulus. Problem (1) originates from, and the importance of (2) is emphasized by, the temporal stimulus structure of experimental tasks in these studies which allows the subject to predict above the chance level the relevant events and, thus, to differentially prepare himself for these in advance (increased non-specific arousal and selective peripheral sensory orientation, the latter often made possible by insufficient control, have possibly been among these changes). Those studies to which these two (and other) remarks do not apply at all or only to an insignificant degree have generally shown no selective evoked-potential changes (or these changes have occurred only with a long latency ('P3' or 'P300') making their interpretation especially uncertain). There is one exception for this general notion, the reasons for and significance of which are dealt with in detail. Finally, the difficulties and inherent limitations of inferring brain events from scalp-recorded evoked-potential data, especially with respect to the important selective-filter hypothesis of selective attention, are extensively discussed and, in the light of these difficulties, some trends for future research proposed.