Accurate and reproducible patient setup is a prerequisite to fractionated radiotherapy. To evaluate the applicability and technical performance of a commercial 3D surface imaging system for repositioning of breast cancer patients, measurements were performed in a rigid anthropomorphic phantom as well as in healthy volunteers. The camera system records a respiration-gated surface model of the imaged object, which may be registered to a previously recorded reference model. A transformation is provided, which may be applied to the treatment couch to correct the setup of the patient. The system showed a high stability and detected pre-defined shifts of phantoms and healthy volunteers with an accuracy of 0.40 +/- 0.26 mm and 1.02 +/- 0.51 mm, respectively (spatial deviation between pre-defined shift and suggested correction). The accuracy of the suggested rotational correction around the vertical axis was always better than 0.3 degrees in phantom measurements and 0.8 degrees in volunteers, respectively. Comparison of the suggested setup correction with that detected by a second and independently operated marker-based optical system provided consistent results. The results demonstrate that the camera system provides highly accurate setup corrections in a phantom and healthy volunteers. The most efficient use of the system for improving the setup accuracy in breast cancer patients has to be investigated in routine patient treatments.