According to D. E. Broadbent's (1958) selective filter theory, people do not process unattended stimuli beyond the analysis of basic physical properties. This theory was later rejected on the basis of numerous findings that people identify irrelevant (and supposedly unattended) stimuli. A careful review of this evidence, however, reveals strong reasons to doubt that these irrelevant stimuli were in fact unattended. This review exposed a clear need for new experiments with tight control over the locus of attention. The authors present 5 such experiments using a priming paradigm. When steps were taken to ensure that irrelevant stimuli were not attended, these stimuli produced no priming effects. Hence, the authors found no evidence that unattended stimuli can be identified. The results support a modern version of Broadbent's selective theory, updated to reflect recent research advances.

Recent progress in the study of attention and performance is discussed, focusing on the nature of attentional control and the effects of practice. Generally speaking, the effects of mental set are proving more pervasive than was previously suspected, whereas automaticity is proving less robust. Stimulus attributes (e.g. onsets, transients) thought to have a "wired-in" ability to capture attention automatically have been shown to capture attention only as a consequence of voluntarily adopted task sets. Recent research suggests that practice does not have as dramatic effects as is commonly believed. While it may turn out that some mental operations are automatized in the strongest sense, this may be uncommon. Recent work on task switching is also described; optimal engagement in a task set is proving to be intimately tied to learning operations triggered by the actual performance of a new task, not merely the anticipation of such performance.

The present study assessed three hypotheses of how practice reduces dual-task interference: Practice teaches participants to efficiently integrate performance of a task pair; practice promotes automatization of individual tasks, allowing the central bottleneck to be bypassed; practice leaves the bottleneck intact but shorter in duration. These hypotheses were tested in two transfer-of-training experiments. Participants received one of three training types (Task 1 only, or Task 2 only, or dual-task), followed by dual-task test sessions. Practice effects in Experiment 1 (Task 1: auditory-vocal; Task 2: visual-manual) were fully explained by the intact bottleneck hypothesis, without task integration or automatization. This hypothesis also accounted well for the majority of participants when the task order was reversed (Experiment 2). In this case, however, there were multiple indicators that several participants had succeeded in eliminating the bottleneck by automatizing one or both tasks. Neither experiment provided any evidence that practice promotes efficient task integration.

Can people learn to perform two tasks at the same time without interference? To answer this question, the authors trained 6 participants for 36 sessions in a Psychological Refractory Period (PRP) experiment, where Task 1 required a speeded vocal response to an auditory stimulus and Task 2 required a speeded manual response to a visual stimulus. The large PRP effect found initially (353 ms in Session 1) shrank to only about 40 ms over the course of practice, disappearing entirely for 1 of the 6 participants. This reduction in the PRP effect with practice is considerably larger than has been previously reported. The obtained pattern of factor interactions between stimulus onset asynchrony and each of three task difficulty manipulations (Task 1 judgment difficulty, Task 2 stimulus contrast, and Task 2 mapping compatibility) supports a postponement (bottleneck) account of dual-task interference, both before and after practice.