Prototypes of personality were investigated in two studies. In study I, clusters of Big-Five-based prototypes were examined using a general population sample of 1908 German adults. Convergent evidence suggested the appropriateness of a five-cluster solution, which corresponds to previously identified temperament based prototypes. In study II, the five-cluster solution was cross-validated in a sample of 256 prisoners. Moreover, it was shown that a population-based approach (using discriminant functions derived from study I) was superior over the traditional sample-based cluster approach (using Ward followed by k-means). The authors argue that future typological research can be sufficiently grounded on a five-prototype conception rather than on a three-prototype conception, and suggest a new and flexible assignment procedure. Copyright © 2006 John Wiley & Sons, Ltd.