Reasons are given for believing that morphological distyly is unlikely to evolve before the incompatibility system. The first problem studied is therefore the evolution of incompatibility of the type found in distyly, with two pollen and two stigma types. It is argued that the first step, starting from self-compatibility, is a mutation to a new pollen type, such that the mutant cannot self-fertilize or fertilize other individuals. Conditions for the spread of such a mutation (assumed recessive) are given. Next, a dominant mutation to a new stigma type, incompatible with the original pollen type but compatible with the new type, is studied. Such a mutation is almost certain to be eliminated if it occurs first, but can spread in a population polymorphic for the pollen mutation provided that it occurs at a linked locus. With tight linkage, this model generates a population with two incompatibility types, one dominant to the other, with only rare recombinant types. It is shown by computer runs that modifiers ...