Record linkage and capture-recapture models were used to estimate the number of cases of human leptospirosis in the health district of Santa Maria, RS in southern Brazil. Twelve months of laboratory, hospital and epidemiological surveillance data were matched by name, age, residence and the month of diagnosis. Only laboratory-confirmed cases were considered. The record linkage revealed more than 20 times more cases than the official estimate for the health district, indicating a leptospirosis epidemic, with an annual incidence of more than 3 per 1,000 inhabitants and a case fatality of 0.37%. Severe cases were predominantly found through hospital records, overlapping to some extent with the epidemiological surveillance data, whereas less severe cases were found almost exclusively through laboratory logs. Different combinations of data sources influenced the detection rate for low versus high severity cases. Based on log-linear capture-recapture models, stratified by case severity and taking into account possible dependencies between the data sources, an insignificant number of cases were missed by all sources.