A recent study (Crosbie & Kelly, 1994) showed that brief externally imposed postfeedback delays improved performance in computerized programmed instruction. The present experiment extended that analysis by using shorter sessions, a more powerful single-subject design, and better timing measures, and by assessing maintenance of training effects. College students completed 40 sets of Holland and Skinner's (1961) programmed text on behavior analysis in a computerized format in a two-component multiple schedule. In the components there was either no postfeedback delay or a 10-s postfeedback delay for each question. On average, delay improved performance by 7% during training, and this advantage was maintained and increased at both posttest (13%) and follow-up (17%). Subjects were satisfied with delay, and session time increased only slightly. Present results show that subjects work faster and more accurately, and have better maintenance when their pace is externally imposed rather than self-selected.