One way to reduce patient set-up errors in radiotherapy is to measure the position during the first N treatment fractions, and to do an unconditional correction of the set-up position once at the (N + 1)th fraction. This strategy is known as the 'no action level' protocol. The question is when to do the correction, i.e. what is the optimum value of N? We determine N by minimizing the expectation value of the total quadratic set-up error taken over all fractions. A central assumption that we make is that there is no time trend in the patient set-up. The result is a simple formula for the value of N, which is proportional to the square root of the total number of fractions, and to the ratio of the execution (delivery) error and preparation error. We also provide a formula for cases where the measurement error is not negligible. For typical cases the optimum value is N = 4. Because the optimum is shallow, the exact choice of N is uncritical.