Selecting the cut-off value to identify infected individuals requires a trade-off between sensitivity and specificity. The standard technique for selecting a cut-off uses results of a population either known or assumed uninfected. Using mixture models, however, it is not necessary to identify an uninfected population. This technique also provides methods to estimate the probability that an individual will be accurately classified. We illustrate this technique in determining cut-off values for the serological diagnosis of human parvovirus B19 infection.