Two critical target stimuli (T1 and T2) were embedded in a stream of white letters shown on a black background, using a rapid serial visual presentation paradigm (RSVP, 100 msec/item). T1 was a red H or S; T2 was an X or a Y. Performance in a two-alternative discrimination on T2 was impaired when processing of T1 was required--a result often called an attentional blink (AB). In previous work, the response in Task1 has been an unspeeded and delayed response at the end of the trial. Three experiments compared performances in Task2 that depended on whether Task1 required an unspeeded delayed response or a speeded immediate response. A larger AB was found when a speeded response was required. Furthermore, in the speeded conditions, faster responses in Task1 were associated with a smaller and shorter AB effect than were slower responses. The results show that manipulations affecting a relatively late stage of processing--response selection--affect the magnitude and duration of the AB phenomenon. A new central inhibition theory is proposed to account for these results. According to this theory, the AB is similar to the psychological refractory period effect and is caused by central postponement of short-term consolidation of T2.