BACKGROUND: The 2001 and 2005 McDonald criteria allow MRI evidence for dissemination in space (DIS) and dissemination in time (DIT) to be used to diagnose multiple sclerosis in patients who present with clinically isolated syndromes (CIS). In 2006, new criteria were proposed in which DIS requires at least one T2 lesion in at least two of four locations (juxtacortical, periventricular, infratentorial, and spinal-cord) and DIT requires a new T2 lesion on a follow-up scan. We applied all three criteria in a large cohort of CIS patients to assess their performance by use of conversion to clinically definite multiple sclerosis (CDMS) as the outcome. METHODS: Patients who had two MRI scans within 12 months of CIS onset were identified in four centres in the Magnims European research network. The specificity and sensitivity of MRI criteria for CDMS after 3 years was assessed in 208 patients. A Cox proportional hazards model was applied in a larger cohort of 282 patients that included all patients irrespective of length of follow-up. FINDINGS: The specificity of all criteria for CDMS was high (2001 McDonald, 91%; 2005 McDonald, 88%; new, 87%). Sensitivity of the new (72%) and 2005 McDonald (60%) criteria were higher than the 2001 McDonald criteria (47%). The Cox proportional hazards model showed a higher conversion risk for all three criteria in those with both DIS and DIT than those with either DIS or DIT alone. When all three criteria were included in the model, only the new criteria had an independent significant effect on conversion risk. INTERPRETATION: The new criteria are simpler than the McDonald criteria without compromising specificity and accuracy. The presence of both DIS and DIT from two MRI scans has a higher specificity and risk for CDMS than either DIS or DIT alone.