Recent applications of three-dimensional treatment planning using non-coplanar treatment angles have demonstrated the potential for improved target dose homogeneity as well as normal tissue sparing for tissues in the neck, notably the parotid gland. Implementation of these highly targeted treatments requires a reliable method for accurate daily reproduction of the treatment position, as patient setup error could lead to significant decreases in target dose and normal tissue sparing. Additionally, the constrained geometry of this anatomic site requires the freedom to deliver treatments from any arbitrary angle in order to maximize the expected benefit. In order to permit accurate setup while maintaining access from most angles, a hybrid immobilization system has been developed, consisting of a custom thermoplastic mask with attachment to a foam cradle shaped to the back of the patient. To evaluate the accuracy of this system, setup errors were measured in 20 patients treated while immobilized with the positioning aide. Two orthogonal film sets consisting of anterior and lateral projections, one set taken at the beginning of treatment and a second 4-5 weeks into therapy, were compared to baseline simulator films or digitally reconstructed radiographs. The average setup deviation in any direction ranged from 0.8 to 1.4 mm and the largest single setup error observed was 4.5 mm. For the film sets taken late in treatment only 3 of the 20 patients required setup adjustment followed by repeat filming to obtain an acceptable film pair. This system has been implemented for routine clinical use since March 1995.