Butterworth (1999) suggested that fingers are important in representing numerosities. Furthermore, scores on a finger gnosis test are a better predictor of numerical performance up to 3 years later than intellectual measures (Marinthe et al., 2001; Noël, 2005). We hypothesised that training in finger differentiation would increase finger gnosis and might also improve numerical performance. Accordingly, 47 first-grade children were selected and divided into 3 groups: children with poor finger gnosis who followed the finger-differentiation training programme (G1), a control-intervention who were trained in story comprehension (G2), and a group with high finger gnosis scores who just continued with normal school lessons (G3). The finger training consisted of 2 weekly sessions of half an hour each, for 8 weeks. Before the training period, children in G3 performed better in finger gnosis and enumeration than children in the two other groups. After the training period this pattern remained for the children in G2 and G3, but the children in G1 were significantly better than those in G2 at finger gnosis, representation of numerosities with fingers, and quantification tasks; they also tended to be better at the processing of Arabic digits. These results indicate that improving finger gnosis in young children is possible and that it can provide a useful support to learning mathematics. Such an approach could be particularly appropriate for children with a developmental Gerstmann syndrome. Theoretically, these results are important because they suggest a functional link between finger gnosis and number skills.