This article proposes a two recursive equation model to correct for endogeneity in latent class Probit models. Concretely, it is assumed that there exists an endogenous and continuous variable defined as a predictor, while unobserved heterogeneity is conceptualized as a vector of parameters that varies across individuals following a discrete distribution. A Maximum Likelihood Estimator is provided to estimate the model parameters based on normally distributed random terms and a free code in R software is provided to carry out the estimation procedure. A small Monte Carlo experiment is carried out to analyze the properties of the estimator. Finally, the estimator is applied to analyze the heterogeneous effects of weight on mental well-being.