Our goal was to evaluate the accuracy of stereotactic technique using MRI in thalamic functional neurosurgery. A phantom study was designed to estimate errors due to MRI distortion. Stereotactic mechanical accuracy was assessed with the Suetens-Gybels-Vandermeulen (SGV) angiographic localiser. Three-dimensional MRI reconstructions of 86 therapeutic lesions were performed. Their co-ordinates were corrected from adjustments based on peroperative electrophysiological data and compared to those planned. MR image distortion (maximum: 1 mm) and chemical shift of petroleum oil-filled localiser rods (2.2 mm) induced an anterior target displacement of 2.6 mm (at a field strength of 1.5 T, frequency encoding bandwidth of 187.7 kHz, on T1-weighted images). The average absolute error of the stereotactic material was 0.7 mm for anteroposterior (AP), 0.5 mm for mediolateral (ML) and 0.8 mm for dorsoventral (DV) co-ordinates (maximal absolute errors: 1.6 mm, 2.2 mm and 1.7 mm, respectively; mean euclidean error: 1 mm). Three-dimensional MRI reconstructions showed an average absolute error of 0.8 mm, 0.9 mm and 1.9 mm in AP, ML and DV co-ordinates, respectively (maximal absolute errors: 2.4 mm, 2.7 mm and 5.7 mm, respectively; mean euclidean error: 2.3 mm). MRI distortion and chemical-shift errors must be determined by a phantom study and then compensated for. The most likely explanation for an average absolute error of 1.9 mm in the DV plane is displacement of the brain under the pressure of the penetrating electrode. When this displacement is corrected for by microelectrode recordings and stimulation data, MRI offers a high degree of accuracy and reliability for thalamic stereotaxy.