To test the accuracy and reproducibility of the tangential breast treatment set-up used in The Netherlands Cancer Institute, a portal imaging study was performed in 12 patients treated for early stage breast cancer. With an on-line electronic portal imaging device (EPID) images were obtained of each patient in several fractions and compared with simulator films and with each other. In five patients, multiple images (on the average 7) per fraction were obtained to evaluate set-up variations due to respiratory movement. The central lung distance (CLD) and other set-up parameters varied within one fraction about 1 mm (1 SD). The average variation of these parameters between various fractions was about 2 mm (1 SD). The differences between simulator and treatment set-up over all patients and all fractions was on the average 2-3 mm for the central beam edge to skin distance and the central lung distance. It can be concluded that the tangential breast treatment set-up is very stable and reproducible and that respiration does not have a significant influence on treatment volume. The EPID appears to be an adequate tool for studies of treatment set-up accuracy like this.