Background: In 2016, the Magnetic Resonance Imaging in Multiple Sclerosis (MAGNIMS) network proposed modifications to the MRI criteria to define dissemination in space (DIS) and time (DIT) for the diagnosis of multiple sclerosis in patients with clinically isolated syndrome (CIS). Changes to the DIS definition included removal of the distinction between symptomatic and asymptomatic lesions, increasing the number of lesions needed to define periventricular involvement to three, combining cortical and juxtacortical lesions, and inclusion of optic nerve evaluation. For DIT, removal of the distinction between symptomatic and asymptomatic lesions was suggested. We compared the performance of the 2010 McDonald and 2016 MAGNIMS criteria for multiple sclerosis diagnosis in a large multicentre cohort of patients with CIS to provide evidence to guide revisions of multiple sclerosis diagnostic criteria. Methods: Brain and spinal cord MRI and optic nerve assessments from patients with typical CIS suggestive of multiple sclerosis done less than 3 months from clinical onset in eight European multiple sclerosis centres were included in this retrospective study. Eligible patients were 16-60 years, and had a first CIS suggestive of CNS demyelination and typical of relapsing-remitting multiple sclerosis, a complete neurological examination, a baseline brain and spinal cord MRI scan obtained less than 3 months from clinical onset, and a follow-up brain scan obtained less than 12 months from CIS onset. We recorded occurrence of a second clinical attack (clinically definite multiple sclerosis) at months 36 and 60. We evaluated MRI criteria performance for DIS, DIT, and DIS plus DIT with a time-dependent receiver operating characteristic curve analysis. Findings: Between June 16, 1995, and Jan 27, 2017, 571 patients with CIS were screened, of whom 368 met all study inclusion criteria. At the last evaluation (median 50·0 months [IQR 27·0-78·4]), 189 (51%) of 368 patients developed clinically definite multiple sclerosis. At 36 months, the two DIS criteria showed high sensitivity (2010 McDonald 0·91 [95% CI 0·85-0·94] and 2016 MAGNIMS 0·93 [0·88-0·96]), similar specificity (0·33 [0·25-0·42] and 0·32 [0·24-0·41]), and similar area under the curve values (AUC; 0·62 [0·57-0·67] and 0·63 [0·58-0·67]). Performance was not affected by inclusion of symptomatic lesions (sensitivity 0·92 [0·87-0·96], specificity 0·31 [0·23-0·40], AUC 0·62 [0·57-0·66]) or cortical lesions (sensitivity 0·92 [0·87-0·95], specificity 0·32 [0·24-0·41], AUC 0·62 [0·57-0·67]). Requirement of three periventricular lesions resulted in slightly lower sensitivity (0·85 [0·78-0·90], slightly higher specificity (0·40 [0·32-0·50], and similar AUC (0·63 [0·57-0·68]). Inclusion of optic nerve evaluation resulted in similar sensitivity (0·92 [0·87-0·96]), and slightly lower specificity (0·26 [0·18-0·34]) and AUC (0·59 [0·55-0·64]). AUC values were also similar for DIT (2010 McDonald 0·61 [0·55-0·67] and 2016 MAGNIMS 0·61 [0·55-0·66]) and DIS plus DIT (0·62 [0·56-0·67] and 0·64 [0·58-0·69]). Interpretation: The 2016 MAGNIMS criteria showed similar accuracy to the 2010 McDonald criteria in predicting the development of clinically definite multiple sclerosis. Inclusion of symptomatic lesions is expected to simplify the clinical use of MRI criteria without reducing accuracy, and our findings suggest that needing three lesions to define periventricular involvement might slightly increase specificity, suggesting that these two factors could be considered during further revisions of multiple sclerosis diagnostic criteria. Funding: UK MS Society, National Institute for Health Research University College London Hospitals Biomedical Research Centre, Dutch MS Research Foundation.