A portal imaging system has been used, in conjunction with a movie measurement technique to measure set-up errors for 15 patients treated with radiotherapy of the pelvis and for 12 patients treated with radiotherapy of the brain. The pelvic patients were treated without fixation devices and the brain patients were treated with individually-moulded plastic shells. As would be expected the brain treatments were found to be more accurate than the pelvic treatments. Results are presented in terms of five error types: random error from treatment to treatment, error between mean treatment position and simulation position, random simulation error, systematic simulator-to-treatment errors and total treatment error. For the brain patients the simulation-to-treatment error predominates and random treatment errors were small (95% ≤ 3 mm, 77% ≤ 1.5 mm). Vector components of the systematic simulation-to-treatment errors were 1-2 mm with maximal random simulation error of ± 5 mm (2 S.D.). There is much interest in the number of verification films necessary to evaluate treatment accuracy. These results indicate that one check film performed at the first treatment is likely to be sufficient for set-up evaluation. For the pelvis the random treatment error is larger (95% ≤ 4.5 mm, 87% ≤ 3 mm). The systematic simulation-to-treatment error is up to 3 mm and the maximal random simulation error is ± 6 mm (2 S.D.). Thus corrections made solely on the basis of a first day check film may not be sufficient for adequate set-up evaluation.