A latent class probit model is developed in which it is assumed that the binary data of a particular subject follow a finite mixture of multivariate Bermoulli distributions. An EM algorithm for fitting the model is described and a Monte Carlo procedure for testing the number of latent classes...