Thirty years ago, in his inaugural article entitled 'The somatic generation of immune recognition', Niels Jerne put forward the hypothesis that the primary antigen (Ag)-receptor repertoire must be restricted towards self-Ags before Ag-mediated selection. The subsequent discovery that Ag receptors are encoded by random rearrangements between discontinuous gene segments was, apparently, at odds with this hypothesis. However, recent findings have begun to reconcile these two concepts. The recombination process is, in fact, relatively precise, exhibiting marked preferences for some gene segments over others, even among members of the same gene family. The result is an intricately patterned primary repertoire that accommodates both sets of predictions, ensuring a balance between the efficiency of selection (requiring limited diversity) and the complexity of the repertoire (requiring maximum diversity).