Computerized record linkage has been used increasingly in epidemiologic studies. We developed a multistage, deterministic matching algorithm using various combinations of key variables. Then, from the records for March 1, 1993, to March 31, 1996, contained in the discharge abstract database of the Canadian Institute for Health Information (CIHI), we examined the relation between length of hospital stay at birth and neonatal readmission. A combined use of province/territory of occurrence, 6-digit postal code of residence, date of birth and sex (step 1) matched 88.5% of 26,629 eligible neonatal readmission records with their birth records. Additional use of institution code and chart number or health card number combined with date of birth and sex (step 2 and step 3) increased the matching rate to 93.0%. Compared with the gold standard, step 1 correctly matched 94.4% of the records. We conclude that this deterministic matching algorithm is a feasible and convenient approach to data linkage for the study of neonatal readmission. The linkage strategy may also be helpful in epidemiologic studies of other short-term events