The European Organization for Research and Treatment of Cancer (EORTC) Radiotherapy Group initiated its mailed thermoluminescence dosimetry (TLD) programme in 1986. The aim of the present study was to evaluate the clinical relevance of variations in beam output detected in the period 1993 to 1996. A total of 140 beam outputs were checked (26 for cobalt-60 units and 114 for linear accelerators) in 35 centres. Clinical dose-response data for tumour control and normal tissue morbidity were used to assess the variation in clinical outcome resulting from variability in beam output. For 75 checked beams with nominal accelerating potentials (n. a.p.) of 6 MV or less the mean ratio, +/- standard deviation (S.D.) of measured to stated output was 1.004+/-0.020. For 65 beams with n. a.p. of 8 MV or more, the ratio was 1.009+/-0.021. Even with this relatively high level of precision, broad distributions of estimated tumour control or normal tissue morbidity were found. In the 10% of the beams with the most pronounced underdosage, the loss in tumour control probability was estimated at 7-8 percentage points. Likewise, in the 10% of the beams with the most pronounced overdosage, the increase in mild/moderate morbidity was 19-22 percentage points. For severe morbidity the same beams raised the estimated incidence of severe complications from 5% to 9-10%. An estimation of the loss of uncomplicated cure probability was about 1% for both high and low energy beams. Sequential mailings considerably improved the uniformity of clinical outcome. We conclude that small deviations in beam output may lead to clinically important variations in outcome. Substantial reductions in the variation between measured and stated output can be achieved by sequential mailings. Mailed TLD checks should be an integral part of a continuously ongoing quality assurance activity in radiotherapy.