The difficulties in a dimensional approach to the description of patient symptomatology, in the case of psychotic illnesses, arise from the fact that many symptoms are of rare incidence, while the more common symptoms tend to be recorded only when they are present to an extent which is thought to be pathological. Nevertheless, variables 1 to 23 in Table I, which may be called the neurotic variables, show considerable variability in both the schizophrenic and affective psychotic samples (see Tables II and IV) and these variables give rise to one clear dimension of variability namely the factor retarded depression, and two other group factors (see Tables III and V). These factors could be defined with some rigour if the neurotic variables had a `lower floor'. We may take the first variable, namely worry, again as an example. Instead of scoring worry of a pathological kind only, a psychiatrist might feel able to extend the scale of this variable to include indications of worry below the pathological level. In some measure he does reach a low floor on other variables, as the distributions of numbers 2 and 11, namely tension and low self-opinion, in Table IV shows. Given lower floors the distributions of the neurotic variables would more closely approach normality, and the factors derived from them could then be defined with greater precision. But the possibility of defining in any acceptable sense dimensions in terms of the psychotic variables, namely 25 to 44 in Table I, is extremely remote. These variables are too limited and restricted in range. In passing, we may recall (Maxwell, 1971) that even if it were possible to describe patients' symptomatology adequately in terms of dimensions this would not contribute to a typology or classification of patients, for all patients would have scores on all dimensions and for each dimension the distribution of scores would tend strongly to normality, with the majority of patients clustering round the mean. In view of the partial success reached by Everitt, Gourlay and Kendell in validating currently used typologies by means of a cluster analysis, and the main conclusion reached in this study that the covariation between the neurotic variables can be accounted for largely by one dominant and a few group factors, Torgerson's suggestion, stated in our introduction, is borne out (if we overlook the fact that it is in part a non-sequitur). But it is not the full story, since rare symptoms of a psychotic type are not amenable to a dimensional description and have a fragmenting effect on a typological system.