We prospectively studied 163 patients referred for MRI of the temporal bone. A presumed diagnosis was made using only one of three sequences: a single thick (12 mm) slice fast T2-sequence, 3D fourier transform constructive interference in steady state (3DFT-CISS) sequence and a gadolinium-enhanced T1-weighted sequence. The visibility of the cochlea, vestibule and superior, lateral and posterior semicircular canals of normal temporal bones was assessed on the T2-weighted images: they were almost always visible (98-100%), with exception of the superior semicircular canal, seen in only 35% of cases. The images were interpreted as abnormal in 34 patients (21%). Using only the fast T2-weighted, 3DFT-CISS and gadolinium-enhanced T1-weighted sequences a presumed false positive diagnosis was made in 5, 1 and 0 cases and a false negative diagnosis in 2, 2 and 4 cases respectively. The overall reliability of the thick-section fast T2-weighted images is limited. This study suggests that a combination of gadolinium-enhanced T1-weighted and 3DFT-CISS images can be considered the gold standard for temporal bone MRI and neither sequence performed separately is as accurate as both together.