The SF-36 was interviewer-administered to 305 subjects at recruitment. Feasibility, content validity and internal consistency were assessed. We tested a priori hypotheses about discriminative, convergent and divergent validity. Interviewer-assisted administration was feasible. The content validity of several domains (Physical Function, Role Physical, Social Function and Role Emotional) was compromised by the irrelevance of some items and response options. Resultant ceiling and floor effects may limit the SF-36?s ability to detect changes over time. The SF-36 was able to discriminate differences between people with: tetraplegia versus paraplegia (in the Physical Function and Physical Composite scores); injuries that were recent ( 4 years) (in the Vitality, Social Function and Mental Health domain and Mental Composite scores), and who were employed versus unemployed (in the Physical Function, Social Function, Mental Health and Mental Composite scores). It was not able to discriminate between groups dichotomised by age, injury completeness or gender. The convergent and divergent validity of all SF-36 domains was as in other populations, except for correlations involving the Physical Function scale which were poor. Internal consistency was similar to that in other populations (Cronbach?s alpha from 0.75 to 0.92); the SF-36 has sufficient precision for population-based and clinical research in spinal cord injury. The SF-36 is useful for comparing the health status of people with spinal cord injury to that of other populations, but supplementation with a disease-specific health status measure may be necessary for trials of interventions in people with spinal cord injuries.