The aim of this study was to identify first admissions in a public hospital discharge file with the greatest possible accuracy. Computerised data linkage was used to link injury events. This involved the use of "internal" data linkage (unduplication) which, in data linkage terms, is equivalent to matching two identical files. Admission status indicators obtained from deterministic and probabilistic linkages were compared with those obtained from a manual review. Small absolute reductions in error were obtained using a probabilistic linkage over a deterministic linkage. However, these reflected large relative reductions in error. A validity check confirmed initial results and discounted against possible bias due to the subjective nature of the probabilistic linking procedure.