Study design: Data are from a multicenter, randomized, double blind clinical trial of acute spinal cord injury. Objectives: To evaluate the prognostic value of magnetic resonance imaging (MRI) for randomized patients in the National Acute Spinal Cord Injury Study 3 (NASCIS). Setting: Sixteen spinal cord injury centers throughout the United States and Canada. Methods: Of 499 patients randomized in NASCIS 3 between December 1991 and September 1995, MRI was electively done on 191 patients within 72 h of injury. Indications of hemorrhage, edema, and contusion were recorded by standard protocol. Neurological impairment as determined by motor function, response to pin prick and light touch was assessed at admission to the participating center and 6 weeks after injury. Change in neurological function was obtained by subtracting the score of each neurological parameter at admission from that measured at 6 weeks. Spinal cord surgery performed within the 3 days after injury was noted. Data were analyzed by: chi square, analysis of variance, multiple logistic regression and linear regression models. Results: Patients with hemorrhage were much more likely to have a complete injury (OR=2.88, 95 Cl 1.32, 6.23); however this association was much reduced when the initial neurological examination was taken into account (AOR=1.43, 95% Cl 0.55, 3.73) and was no longer a signi®cant predictor of injury. MRI evidence of cord edema was the strongest predictor of reduced improvement in motor function (73.34 points, P=0.06) and light touch sensation (73.41 points, P=0.05) at 6 weeks. Conclusions: Cord hemorrhage, contusion, and edema on MRI were not associated with diagnosis of a complete cord injury after neurological assessment from the initial clinical examination was taken into account. Prediction of a worse 6 week neurological status was weakly associated with the presence of edema diagnosed by MRI. As MRI technology improves, these diagnostic and predictive capabilities need to be re-assessed.