OBJECTIVE To determine whether the clinical implementation of an electronic portal imaging device can improve the precision of daily external beam radiotherapy. METHODS In 1991, an electronic portal imaging device was installed on a dual energy linear accelerator in our clinic. After training the radiotherapy technologists in the acquisition and evaluation of portal images, we performed a randomized study to determine whether online observation, interruption, and intervention would result in more precise daily setup. The patients were randomized to one of two groups: those whose treatments were actively monitored by the radiotherapy technologists and those that were imaged but not monitored. The treating technologists were instructed to correct the following treatment errors: (a) field placement error (FPE) > 1 cm; (b) incorrect block; (c) incorrect collimator setting; (d) absent customized block. Time of treatment delivery was recorded by our patient tracking and billing computers and compared to a matched set of patients not participating in the study. After the patients radiation therapy course was completed, an offline analysis of the patient setup error was planned. RESULTS Thirty-two patients were treated to 34 anatomical sites in this study. In 893 treatment sessions, 1,873 fields were treated (1,089 fields monitored and 794 fields unmonitored). Ninety percent of the treated fields had at least one image stored for offline analysis. Eighty-seven percent of these images were analyzed offline. Of the 1,011 fields imaged in the monitored arm, only 14 (1.4%) had an intervention recorded by the technologist. Despite infrequent online intervention, offline analysis demonstrated that the incidence of FPE > 10 mm in the monitored and unmonitored groups was 56 out of 881 (6.1%) and 95 out of 595 (11.2%), respectively; p < 0.01. A significant reduction in the incidence of FPE > 10 mm was confined to the pelvic fields. The time to treat patients in this study was 10.78 min (monitored) and 10.10 min (unmonitored). Features that were identified that prevented the technologists from recognizing more errors online include poor image quality inherent to the portal imaging device used in this study, artifacts on the portal images related to table supports, and small field size lacking sufficient anatomical detail to detect FPEs. Furthermore, tools to objectively evaluate a portal image for the presence of field placement error were lacking. These include magnification factor corrections between the simulation of portal image, online measurement tools, image enhancement tools, and image registration algorithms. CONCLUSIONS The use of an electronic portal imaging device in our clinic has been implemented without a significant increase in patient treatment time. Online intervention and correction of patient positioning occurred rarely, despite FPEs of > 10 mm being present in more than 10% of the treated fields. A significant reduction in FPEs exceeding 10 mm was made in the group of patients receiving pelvic radiotherapy. It is likely that this improvement was made secondarily to a decrease in systematic error and not because of online interventions. More significant improvements in portal image quality and the availability of online image registration tools are required before substantial improvements can be made in patient positioning with online portal imaging.