Building on prior work, the current study evaluated whether connectionist models can account for the distance and size effects in adults and the development of the distance effect in children. A family of models was constructed by orthogonally varying training environment (naturalistic versus non-naturalistic) and number representation (one-to-one versus magnitude). The ability of the models to account for the adult distance and size effects depended critically on a naturalistic training environment but was relatively independent of number representation. With respect to the developmental data, the naturalistic/one-to-one model provided a good account of response times and errors. The relation between the current models and prior models and avenues for future exploration are discussed.