The characterization of a mating system critically depends on the level at which mating is considered and on the specification of the mating reference (frequencies of types among the potential mates). Mating preferences among the types present in a population are then defined in terms of conditional mating frequencies (mating norms) and mating references, and the mating success of a type is derived to be the frequency with which individuals of this type participate in all matings, divided by their frequency prior to mating. Mating success is thus closely related to relative Darwinian fitness. This conceptual framework allows the resolution of the stochastic problems inherent in the conventional definition of random mating and results in increased transparency of the notions of sexual selection and assortative mating. In particular, sexual selection turns out to be a system of mating in which one (the selected) sex mates at random with respect to the trait under consideration and the other does not. Mating is shown to be assortative if the mating frequencies deviate from their corresponding product structure, which is a result of differences in patterns of mating preference between types of the same sex. The given definitions lead to methods of analysis which are readily amenable to application in any species. This is demonstrated for a published data set. Some suggestions for the design of experimental studies are presented.