The development of numerical abilities was examined in three groups of 5 year-olds: one including 13 children accomplishing a numerical training in pencil-and-paper format (EG1); another group including 21 children accomplished a homologous training in computerized format; the remaining 24 children were assigned to the control group (CG). The participants were assessed at three successive times (t0, t1 and t2) with a battery of validated tests assessing numerical abilities and fluid intelligence. At times t1 and t2 we found differences between experimental groups and CG, while the children's abilities in the two experimental groups were similar. We underline the crucial role of pre-syntactical and counting dimensions, accounting for a distinction between the experimental groups and control. Results are discussed with reference to the relevance for training activities of the presentation format (pencil-and-paper versus computer-assisted). Pragmatical and practical implications are also considered. [For the full proceedings, see ED562093.]