It is natural for psychiatrists and others dealing with mental patients to enquire what the prognosis is for a patient showing certain symptoms or syndromes of symptoms—taking age and the history of the patient's illness into account. The very fact that the symptoms generally form syndromes points to their interdependence to a greater or lesser degree, but this interdependence may not be a straightforward linear relationship such as is revealed by an ordinary correlation coefficient for the symptoms may interact with each other in a complicated way. When this is the case the problem of prediction is difficult from a statistical as well as from a psychiatric point of view. In this paper an effort is made to clarify the statistical problem by describing briefly and by comparing two statistical models which may be used for predictive purposes. The models are introduced by means of an example using data for patients who have attended this hospital. For reasons of clarity the number of variables included in the study is a bare minimum.