Résumé : We have evaluated the fluoroscopic on-line portal imaging (OPI) system developed by Siemens (Beamview-1®, Concord, CA, U.S.A.) in routine clinical radiotherapy, involving the treatment of 883 fields (559 patient set-ups for treatment) on 21 patients. The image was typically generated by delivering 10 monitor units when used in single exposure or 1-2 monitor units on a large open field followed by 8-10 monitor units on the actual field when double exposure was used. Comparison between the portal image and the simulator film was done by eye. A region of tolerance was drawn on the simulator film and the field edges on the portal image had to project within this region. If this criterion was not met, adjustments followed by verification portal images were done before the remaining field dose was delivered. If possible, these adjustments were performed by moving the patient couch by remote control. The image quality was insufficient for evaluation in 7 5 883 (8.5 %) fields. The abovementioned criterion was not met in 95 808 (11.8 %) of the evaluable fields (26/559 patient set-ups were not evaluable). Of the 533 evaluable patient set-ups, 92 had to be adjusted (17.2%) including three (pelvic irradiations) set-ups that were adjusted on both field irradiated during the same radiotherapy session. In one case an incorrect tray (with wrong blocks) was detected and replaced. In one case (a 5.5 × 6.0 cm rectangular larynx field) the x and y axis of the field were interswitched. In one case incorrect focusing of a block was shown by the portal image. To make adjustments, the couch longitudinal position was changed 20 times (range - 10 to + 15 mm). The lateral position was changed 73 times (range - 15 to + 16 mm). The height position was changes 6 times (range - 7 to + 6 mm). Diaphragma rotation changes were performed 5 times (1°). The fraction of treatment time that was related to the use of OPI was 30.7% median (mean 32.4%, S.D. 14.1%). The range was 4.1 to 78.6%. On the basis of calculations assuming no OPI would have been used, field treatment time was increased by a median of 44.2% (mean 55.8%; S.D. 41.2%) by using OPI. The fraction of monitor units (fraction of the dose) to generate a satisfactory image was 10% median. Errors of field set-up were most often detected on pelvic fields (eight patients, 304 fields) and corrections were made on 79 284 (27.8%) evaluable fields [From 147 patient set-ups with at least one evaluable field, 76 were changed (51.7%); three set-ups were changed on both the anterior and the posterior field of the same session]. Lateral shifts of the bony pelvis up to 1.6 cm were observed even when all skin marks (field axis and edges, lateral and longitudinal lasers) matched perfectly. © 1992.