This paper describes an objective, quantifiable technique for assaying the degree of severity of spinal cord injury. Twenty-one rats underwent a C7-T1 laminectomy: 12 received a C8 spinal cord clip compression injury with forces of either 2.3, 16.9 or 53.0 g; 4 had cord transection at C8, and 5 had no cord lesion. Postoperative clinical neurological assessment was performed by the inclined plane method. At 4 weeks, the spinal cord-injured rats underwent a T10 transection and insertion of a Gelfoam pledget impregnated with horseradish peroxidase (HRP). HRP was similarly administered to 9 normal rats. Longitudinal sections of the spinal cord encompassing the injury site were stained with Luxol fast blue, and coronal sections from the cerebrum and midbrain were processed for HRP reactivity with tetramethylbenzidine. Labelled corticospinal and rubrospinal neurons were counted in every 6th section to derive a cortical score (CS) and a red nucleus score (RNS) for each animal. The CS reflected the extent of the pathological changes at the site of cord injury and the ln CS decreased linearly with increasing injury severity (P less than 0.0001). In contrast, the RNS was only reduced in animals with severe (53.0 g) cord injuries (P less than 0.0006). The degree of preservation of the dorsal columns including the corticospinal tracts at the injury site correlated with the CS, whereas the RNS was related to the degree of preservation of the lateral columns. Counts of rubrospinal neurons, but not corticospinal neurons, correlated closely (r = 0.96, P less than 0.001) with the inclined plane results, suggesting the importance of non-pyramidal tracts in controlling gross motor function. Thus, counting corticospinal and rubrospinal neurons is an objective, reliable test of the severity of experimental spinal cord injury.