Using mathematical models that combine population genetic and epidemiological processes, we resolve the paradox that many important pathogens appear to persist as discrete strains despite the constant exchange of genetic material. We show that dominant polymorphic determinants (that is, those that elicit the most effective immune responses) will be organized into nonoverlapping combinations as a result of selection by the host immune system, thereby defining a set of discrete independently transmitted strains. By analysing 222 isolates of Neisseria meningitidis, we show that two highly polymorphic epitopes of the outer membrane protein PorA exist in nonoverlapping combinations as predicted by this general framework. The model indicates that dominant polymorphic determinants will be in linkage disequilibrium, despite frequent genetic exchange, even though they may be encoded by several unlinked genes. This suggests that the detection of nonrandom associations between epitope regions can be employed as a novel strategem for identifying dominant polymorphic antigens.