An algorithm was developed to statistically predict ischemic tissue fate on a pixel-by-pixel basis. Quantitative high-resolution (200 x 200 microm) cerebral blood flow (CBF) and apparent diffusion coefficient (ADC) were measured on acute stroke rats subjected to permanent middle cerebral artery occlusion and an automated clustering (ISODATA) technique was used to classify ischemic tissue types. Probability and probability density profiles were derived from a training data set (n=6) and probability maps of risk of subsequent infarction were computed in another group of animals (n=6) as ischemia progressed. Predictions were applied to overall tissue fate. Performance measures (sensitivity, specificity, and receiver operating characteristic) showed that prediction made based on combined ADC+CBF data outperformed those based on ADC or CBF data alone. At the optimal operating points, combined ADC+CBF predicted tissue infarction with 86%+/-4% sensitivity and 89%+/-6% specificity. More importantly, probability of infarct (P(I)) for different ISODATA-derived ischemic tissue types were also computed: (1) For the 'normal' cluster in the ischemic right hemisphere, P(I) based on combined ADC+CBF data (P(I)[ADC+CBF]) accurately reflected tissue fate, whereas P(I)[ADC] and P(I)[CBF] overestimated infarct probability. (2) For the 'perfusion-diffusion mismatch' cluster, P(I)[ADC+CBF] accurately predicted tissue fate, whereas P(I)[ADC] underestimated and P(I)[CBF] overestimated infarct probability. (3) For the core cluster, P(I)[ADC+CBF], P(I)[ADC], and P(I)[CBF] prediction were high and similar ( approximately 90%). This study shows an algorithm to statistically predict overall, normal, ischemic core, and 'penumbral' tissue fate using early quantitative perfusion and diffusion information. It is suggested that this approach can be applied to stroke patients in a computationally inexpensive manner.