The development of the immune repertoire during neonatal life involves a strong selection process among different clones. We investigate the hypothesis that repertoire selection is carried out during early life by the immune network. There are at least two processes in repertoire selection: clonal expansion and recruitment of clones by the bone marrow. Because both processes occur on time scales of the order of a few days, we argue that both have to be modeled. In a previous differential equation model (De Boer & Perelson, 1991), studied by numerical integration, both clonal expansion and recruitment were present but the rate of recruitment was kept low due to limitations in computational resources.Here we present a new model based upon a two-dimensional shape space. The model is defined as an asynchronous cellular automaton (CA). In the CA model we vary (1) the rate of recruitment and (2) the specificity of the lymphocyte receptors. The networks attain an equilibrium in which the size of the repertoire remains fixed. However, the equilibrium repertoire size increases when the recruitment rate or the receptor specificity is increased. The number of functional idiotypic interactions per clone, i.e., the connectivity, is less dependent on either the receptor specificity or the recruitment rate. These observations confirm the results of our previous study. The CA model contributes to our understanding of pattern formation in immune network models because of its straightforward visualization. Using it we show that the randomness involved in lymphocyte recruitment may play a role in selecting the clones in the actual repertoire.