OBJECTIVE: MRI at 3 T is said to be more accurate than 1.5 T MR, but costs and other practical differences mean that it is unclear which to use. METHODS: We systematically reviewed studies comparing diagnostic accuracy at 3 T with 1.5 T. We searched MEDLINE, EMBASE and other sources from 1 January 2000 to 22 October 2010 for studies comparing diagnostic accuracy at 1.5 and 3 T in human neuroimaging. We extracted data on methodology, quality criteria, technical factors, subjects, signal-to-noise, diagnostic accuracy and errors according to QUADAS and STARD criteria. RESULTS: Amongst 150 studies (4,500 subjects), most were tiny, compared old 1.5 T with new 3 T technology, and only 22 (15 %) described diagnostic accuracy. The 3 T images were often described as “crisper”, but we found little evidence of improved diagnosis. Improvements were limited to research applications [functional MRI (fMRI), spectroscopy, automated lesion detection]. Theoretical doubling of the signal-to-noise ratio was not confirmed, mostly being 25 %. Artefacts were worse and acquisitions took slightly longer at 3 T. CONCLUSION: Objective evidence to guide MRI purchasing decisions and routine diagnostic use is lacking. Rigorous evaluation accuracy and practicalities of diagnostic imaging technologies should be the routine, as for pharmacological interventions, to improve effectiveness of healthcare. KEY POINTS : * Higher field strength MRI may improve image quality and diagnostic accuracy. * There are few direct comparisons of 1.5 and 3 T MRI. * Theoretical doubling of the signal-to-noise ratio in practice was only 25 %. * Objective evidence of improved routine clinical diagnosis is lacking. * Other aspects of technology improved images more than field strength.