OBJECT: The authors sought to evaluate and improve the geometrical accuracy of a 3-tesla magnetic resonance (MR) imaging unit used for Gamma Knife surgery (GKS). METHODS: To evaluate the geometrical accuracy of a Siemens Magnetom Trio 3-tesla MR imaging unit, two phantoms were used. Both phantoms were imaged with computed tomography (CT), a 1.5-tesla MR imaging unit (Siemens Avanto), and the 3-tesla MR imaging unit. A pair of orthogonal films was obtained with a radiotherapy simulator to validate the spatial coordinates of the marker positions determined with CT. The coordinates of the markers were measured using the GammaPlan treatment planning software. Magnetic resonance imaing was performed using three-dimensional (3D) magnetization-prepared rapid acquisition gradient echo (MPRAGE) and fast low-angle shot sequence (FLASH) pulse sequences. The voxel size was 1 x 1 x 1 mm3. CONCLUSIONS: The root-mean-square error of MR images was 2 +/- 0.73 mm for 3D MPRAGE. The error was reduced to 1.5 +/- 0.64 mm for FLASH. The errors were decreased further by applying an image distortion correction method (the field-of-view filter) to the images acquired with FLASH. The mean errors were 1.3 +/- 0.39 mm and 1.5 +/- 0.77 mm for the two phantoms. The errors increased from 1 mm to 3.1 mm as the measurement points approached the caudal edge of the head coil (larger z value). Proper selection of a pulse sequence together with a geometrical distortion correction improved the geometrical accuracy of MR images. However, further study is needed to increase the geometrical accuracy of 3-tesla MR imaging units for radiosurgical applications.