Image co-registration is used in frameless gamma knife radiosurgery (GKSRS) to assign a stereotactic coordinate system and verify patient setup before irradiation. The accuracy of co-registration with cone beam computed tomography (CBCT) images of a Gamma Knife IconTM (GK Icon) was assessed, and the effects of the region of co-registration (ROC) were studied. CBCT-to-CBCT co-registration is used for patient setup verification, and its accuracy was examined by co-registering CBCT images taken at various configurations with a reference CBCT series. The accuracy of stereotactic coordinate assignment was investigated by co-registering stereotactic CT images with CBCT images taken at various configurations. An anthropomorphic phantom was used, and the coordinates of fifteen landmarks inside the phantom were measured. The co-registration accuracy between stereotactic magnetic resonance (MR) and CBCT images was evaluated using images from forty-one patients. The positions of the anterior and posterior commissures were measured in both a fiducial marker-based system and a co-registered system. To assess the effects of MR image distortions, co-registration was performed with four different ranges, and the accuracy of the results was compared. Co-registration between CBCT images gave a mean three-dimensional deviation of 0.2 ± 0.1 mm. The co-registration of stereotactic CT images with CBCT images produced a mean deviation of 0.5 ± 0.2 mm. The co-registration of MR images with CBCT images resulted in the smallest three-dimensional difference (0.8 ± 0.3 mm) when a co-registration region covering the skull base area was applied. The image co-registration errors in frameless GKSRS were similar to the imaging errors of frame-based GKSRS. The lower portion of the patient's head, including the base of the skull, is recommended for the ROC.