This study aimed to explore the heterogeneity of paranoid psychosis with onset in late life by using cognitive factors in a centroid method of cluster analysis. Forty-seven subjects were allocated to two different clusters, the first with 24 (51·1 %) and the second with 23 (48·9 %) patients. Their cognitive attainment was evaluated against the performance of 33 elderly controls, all groups being matched for age, sex, and the numbers of years of education. Patients in cluster 2 showed a pattern of widespread cognitive impairment, which involved general measures of cognitive performance (MMSE, CAMCOG, WAIS-R verbal and performance scores), memory (digit and spatial span, delayed matching-to-sample, recognition memory for words and faces), and executive functions (verbal fluency, extra and intra-dimensional shift ability, spatial working memory, and planning). In contrast, patients in cluster 1 were only impaired on their extra-dimensional set shift and planning abilities, suggesting a more specific and restricted executive functioning deficit. We also analysed the impact that the use of antipsychotic medication could have had on patients' cognitive performance, which was shown to be negligible. In addition, there was no difference between the clusters with regard to the number of patients using neuroleptics, suggesting that the medication was unlikely to have introduced a performance bias in the two patient clusters. The validity of the subdivision of these patients into two separate groups was further supported by other clinical findings. Patients in cluster 1 exhibited more severe psychotic symptoms, as measured by the SAPS, than their counterparts in cluster 2, and were also more likely to display first-rank symptoms of Schneider. Conversely, cluster 2 membership was strongly associated with the presence of neurological signs and negative symptoms. We suggest that psychotic states arising in late life are a heterogeneous condition that may be best divided in two: ‘type A’, including patients with a wide range of psychotic symptoms, mild increase in the frequency of neurological signs, and cognitive deficits restricted to executive functions, and ‘type B’, which includes patients with less complex psychotic symptoms associated with a marked increase in the frequency of neurological signs and generalized cognitive impairment. The basis for this subdivision and the prospect for future studies are discussed.