We are implementing the use of magnetic resonance (MR) images for head and neck radiotherapy planning, which involves their registration with computed tomography (CT). The quality assurance (QA) of the registration process was an initial step of this program. A phantom was built, and appropriate materials were identified to produce clinically relevant MR T1 and T2 contrast for its constituent "anatomy". We performed a characterization of the distortion detectable within our phantom. Finally, we assessed the accuracy of image registration by contouring structures in the registered/fused data sets using the treatment planning system. Each structure was contoured using each modality, in turn, blind of the other. The position, area, and perimeter of each structure were assessed as a measure of accuracy of the entire image registration process. Distortion effects in the MR image were shown to be minimized by choosing a suitable (3 +/- 30 kHz) receiver bandwidth. Remaining distortion was deemed clinically acceptable within +/-15 cm of the magnetic field isocenter. A coefficient of agreement (A) analysis gave values to be within 9% of unity, where A = square root(RaRp) and Ra/p is the ratio of the area/perimeter of a particular structure on CT to that on MR. The center of each structure of interest agreed to within 1.8 mm. A QA process has been developed to assess the accuracy of using multimodality image registration in the planning of radiotherapy for the head and neck; we believe its introduction is feasible and safe.