STUDY DESIGN: Videotape rating by independent viewers. OBJECTIVE: To determine the test-retest reliability of the Donovan spinal cord injury (SCI) pain classification scheme. SETTING: Rehabilitation Centre, Alabama, USA. METHODS: A total of 28 individuals with SCI reported 60 pain sites. A structured interview and physical exam were used to illicit information to classify each pain site according to the Donovan criteria. All structured interviews and exams were videotaped. Three independent raters viewed the videotapes on two occasions, separated by a 3-month interval, and classified each pain site using the Donovan pain classification scheme. RESULTS: Considering all three raters together, 78% of the pain sites were consistently classified from one period to the next. Within each rater, consistent classification ranged from 67 to 83%. However, inter-rater agreement for the classification of each pain site into the various types of pain was low for both periods (about 50-60%). CONCLUSIONS: Pain classification within each rater generally showed adequate test-retest reliability when using the Donovan SCI pain classification scheme. However, reliability estimates of agreement across raters highlight the ongoing need to exam and improve the psychometric characteristics of the various pain classification schemes.