Noting the importance of measuring the pure effects of various disequilibrium parameters in a randomly mating population, the author suggests a class of log-linear models together with methods which enable one to choose the model which best fits the data. More precisely, a log-linear model including only one- and two-factor interactions is suggested, the fit of such a model being assessed by estimating the expected frequencies of all possible types under that model. These estimates are also used to select the most appropriate model, to test hypotheses and to estimate the parameters. Several examples are presented and possible extensions suggested.