Selective attention effects on reflexes and evoked potentials are reviewed with the aim of evaluating three theories regarding sensory automaticity. (a) The peripheral-gating theory, which assumes that ignored stimuli can be filtered out soon after transduction, was tentatively rejected because neither auditory-nerve nor retinal potentials are reliably affected by attention. (b) At the other extreme, the assumption that sensory analyses are obligatory and cannot benefit from attentional resources (i.e., strong-automaticity theory) was also rejected, because longer latency componets were found to be modifiable by attention. (c) An intermediate theory provides the best fit to present electrophysiological data. The earliest sensory analyses are assumed to be strongly automatic and then, at forebrain levels, there is a transition from strong to weak automaticity (i.e., analyses are obligatory but modifiable by attention). This transition can begin as early as about 15 ms for audition and about 80 ms for vision.