In the serodiagnosis of tropical infectious diseases, cut-off values are often established by using sera from individuals living under moderate climatic conditions, not exposed to the risk of infection (non-endemic controls). This approach guarantees the disease-free status of the individuals within that control group but leads to an assembly of samples which are not representative for the disease-free individuals of the target population (selection bias). Using data from an epidemiological study of Trypanosoma evansi infection in dogs, two alternative methods to construct cut-off values for a T. evansi antibody ELISA are described which are solely based on a distribution analysis of the data from the endemic animals. By cluster analysis these data could be divided into 'high', 'intermediate' and 'low responders'. High responders could also be identified by using the computer-assisted analysis of mixtures (C.A.MAN). Conventional cut-offs were calculated from a group of non-endemic individuals. A receiver operating characteristic (ROC) analysis was performed to demonstrate the impact of the choice of cut-offs on the test specificity and on the estimated seroprevalence among the endemic population. The data indicate that distribution analysis, especially the mixture analysis (C.A.MAN), are valuable tools for the unbiased estimation of seroprevalence when representative negative controls are not available.