In the last decades criticism on existing psychiatric classifications has steadily grown. From a theoretical point of view three major objections were raised. Inter-rater reliability is too low, patients within diagnostic classes are heterogeneous rather than homogeneous with respect to characteristics of illness, and predictive validity is insufficient. It is conjectured that these deficiencies stem from the fact that most of the classifications reflect only a limited number of archetypes of psychiatric illnesses, which correspond insufficiently with clinical reality. Instead, classifications should be built on a wide range of data, collected on an elementary level. These bits of information should subsequently be combined in prototypes, using a polythetic rather than a monothetic approach. Since the (probabilistic rather than deterministic) ways of combining will be dependent upon the objectives, this approach will result in a series of classifications, all of which are based upon the same basic information. A multi-axial system for data-collection is proposed. It is felt that this prototypical core-classification may be worthwhile to clinical practice. It will, however, be primarily useful in the fields of psychiatric research and evaluation of mental health care systems.