The purpose of the present work was the construction of statistical models which allow the prediction of future high caries increments on the basis of the state of primary teeth and first molars taking into account both precavity lesions and DF experience. A child was considered to be experiencing 'high caries increment' when it had at least four new lesions (new DFS) 4 years after the first examination. Two data sets of children aged 7-10 years were analysed: (1) 803 children examined in 1980 and in 1984. The proportion of children with 'high caries increment' was 25.0%. (2) 477 children examined in 1984 and in 1988. In this group, the proportion of children with high caries increment was 16.6%. In the first group with higher prevalence of caries, sensitivity and specificity were found to be on average at 70%. In the second group, sensitivity and specificity were 77% on average. This results shows that caries prediction may be at least as successful when a 'refined' set of clinical data is used instead of a less extensive set of caries data complemented by salivary buffering capacity and microbiological data.