OBJECTIVE: To determine bias due to loss of participants (attrition bias) in a prospective cohort study. DESIGN: A multi-centre prospective cohort study. SUBJECTS: A total of 225 individuals with a spinal cord injury from 8 Dutch rehabilitation centres. METHODS: Participants were considered non-participants when no information was collected at the measurement one year after discharge from inpatient rehabilitation. Using bivariate tests participants and non-participants were compared regarding personal, lesion, function and functional characteristics determined at the beginning of inpatient rehabilitation and at discharge. A logistic regression was performed to determine which characteristics predict participation at one year after discharge. RESULTS: Of the participants at the start of the study, 31% (n = 69) did not perform the tests one year after discharge from inpatient rehabilitation. Variables associated with study participation one year after discharge were: higher level of education, higher well-being score at the start of rehabilitation, and a shorter length of stay in hospital and rehabilitation centre at discharge of inpatient rehabilitation. CONCLUSION: Selective attrition in the longitudinal study might have led to an over-estimation of some of the results of the measurement one year after discharge from inpatient rehabilitation.