Objective: Psychosis spectrum disorders are associated with cerebral changes, but the prognostic value and clinical utility of these findings are unclear. Here, we applied a multivariate statistical model to examine the predictive accuracy of global white matter fractional anisotropy (FA) for transition to psychosis in individuals at ultra-high risk for psychosis (UHR). Methods: 110 UHR individuals underwent 3 Tesla diffusion-weighted imaging and clinical assessments at baseline, and after 6 and 12 months. Using logistic regression, we examined the reliability of global FA at baseline as a predictor for psychosis transition after 12 months. We tested the predictive accuracy, sensitivity and specificity of global FA in a multivariate prediction model accounting for potential confounders to FA (head motion in scanner, age, gender, antipsychotic medication, parental socioeconomic status and activity level). In secondary analyses, we tested FA as a predictor of clinical symptoms and functional level using multivariate linear regression. Results: Ten UHR individuals had transitioned to psychosis after 12 months (9%). The model reliably predicted transition at 12 months (χ² = 17.595, p = 0.040), accounted for 15–33% of the variance in transition outcome with a sensitivity of 0.70, a specificity of 0.88 and AUC of 0.87. Global FA predicted level of UHR symptoms (R² = 0.055, F = 6.084, p = 0.016) and functional level (R² = 0.040, F = 4.57, p = 0.036) at 6 months, but not at 12 months. Conclusion: Global FA provided prognostic information on clinical outcome and symptom course of UHR individuals. Our findings suggest that the application of prediction models including neuroimaging data can inform clinical management on risk for psychosis transition.