Background: The risk of multiple sclerosis (MS) is dependent on multiple variables, including geographical location. There is increasing interest in the early recognition and treatment of MS in children. Method: Using univariate and multivariate analysis, we determined the clinical and radiological features that were predictive of MS in 88 children from New South Wales, Australia, with a first acute demyelinating syndrome (ADS) who were followed for a minimum of one year. We tested the McDonald, KIDMUS, Callen and Verhey MRI criteria for paediatric MS. Results: After a mean follow-up of 5.2 years, 13/88 (15%) of children had MS. Using multivariate analysis, preceding infection was protective of MS, and corpus callosal lesions, the combined presence of both well and poorly demarcated lesions, and contrast-enhancing lesions on MRI were predictive of MS. The sensitivity and specificity of the respective radiological criteria were McDonald 2005 (69%, 68%), McDonald 2010 (58%, 95%), KIDMUS (8%, 100%), Callen (69%, 85%) and Verhey (62%, 84%). When McDonald 2010 criteria were applied to baseline and serial scans, the sensitivity and specificity was 91% and 93%. Conclusion: Despite the long follow-up, the risk of MS appears lower in New South Wales children compared to previously reported cohorts. Radiological features are more predictive than clinical features in predicting MS. The McDonald 2010 criteria performed well although the dissemination in time criteria on baseline scans is difficult to apply to children with encephalopathy.