Three experiments explored the way in which delay in the response of the system affects the user's performance in solving problems with an on-line computing service. In each experiment there were four subjects under delay conditions of about 1 sec. to 100 sec. The on-line computing service was the Lincoln Reckoner. As expected, the average time the user required to complete a task increased as the response-delay increased, and the rate at which he demanded service declined as the delay increased. The relation of net completion time (time to complete the task, minus the time during which the user was waiting for a response) to response delay depended on the type of task. In the more realistic experiments, the net completion time increased with delay (suggesting that long delays are distracting). The number of outputs (i.e., displays or type-outs) per task was also considered. The main conclusion is that controlled experiments of this kind are feasible and can be used as the basis for design of on-line computing services.