A probit analysis method is presented for the analysis of preference data from a behavioural assay of animals, where responses were observed as animals' choice between the treatment and control sides or areas. Standard probit models designed for all or nothing phenomena are not applicable to preference assays, because a log dose metameter only indirectly correlates with the choice. The present method enabled analysis by regressing the probit transformation of the proportion of net responders, expressed in a preference model equation, on a log dose metameter. The computer program includes the maximization routine of the log-likelihood function, a test of homogeneity and a calculation of potency with fiducial limits. After the explanation of the basic model, as well as generalized regression models including the parameters of natural preference and immunity, a computer output demonstrated the analysis of an olfactometer assay on two cockroach attractants by reporting statistics. By comparing differently generalized models for fitness to the data, a preferable experimental design was formulated.