The present study analyses the strategies adopted by normal subjects when they are asked to make two separate movements as rapidly as possible one after the other. Five subjects performed the following sequential movements in their own time. 1) Squeeze an isometric force transducer between fingers and thumb to a force of 30 N and then flex the elbow of the same arm through 15°. 2) Squeeze the transducer with one hand and then flex the elbow of the other arm. 3) Perform an isotonic opposition of finger and thumb and then flex the elbow of the same arm. 4) First flex the elbow through 15, 30 or 45° and then squeeze the transducer. 5) Flex and then extend the elbow as rapidly as possible. In tasks 1–4 there was no correlation between the times taken to complete the two separate components of the sequence. Because of this we suggest that the two movements remained under the control of two separate motor programmes. In contrast, in task 5, the times taken for the two components were correlated and hence we suggest that in this case a single programme was used to perform the sequence. In tasks 1–3, in which the mean duration of the first movement was some 135–162 ms, there was a mean pause of about 85 ms before the start of the second movement. Subjects tended to chose a minimum inter-onset latency between the start of the first and the start of the second movement of a sequence of some 230 ms. The reason for this appeared to be that if subjects were encouraged to decrease their interonset latencies to less than 200 ms, the speed of the second movement decreased sharply. However, if the duration of the first movement was prolonged as in task 4, the second movement could be delayed, although there now was little or no pause between the two movements. We conclude that when a single motor programme is run, it is followed by a “relative refractory period”. If a second programme is run within this period, it cannot be executed without loss of speed. Switching from one motor programme to another is achieved with an optimal minimum delay of 200 ms. Sequential movements which are controlled by a single programme do not share this limitation.