There is a response selection bottleneck that is responsible for dual-task interference. How the response selection bottleneck operates was addressed in three dual-task experiments. The overlap between two tasks (as indexed by the stimulus onset asynchrony [SOA]) was systematically manipulated, and both reaction time and electrodermal activity were measured. In addition, each experiment also manipulated some aspect of the difficulty of either task. Both increasing task overlap by reducing SOA and increasing the difficulty of either task lengthened reaction times. Electrodermal response was strongly affected by task difficulty but was only weakly affected by SOA, and in a different manner from reaction time. A fourth experiment found that the subjectively perceived difficulty of a dual-task trial was affected both by task difficulty and by SOA, but in different ways than electrodermal activity. Overall, the results were not consistent with a response selection bottleneck that involves processes of voluntary, executive attention. Instead, the results converge with findings from neural network modeling to suggest that the delay of one task while another is being processed reflects the operation of a routing mechanism that can process only one stream of information for action at a time and of a passive, structural store that temporarily holds information for the delayed task. The results suggest that conventional blocked or event-related neuroimaging designs may be inadequate to identify the mechanism of operation of the response selection bottleneck.