The authors consider the estimation of a sample selection model that exhibits spatial autoregressive errors (SAE). Their methodology is motivated by a two-step strategy where in the first step they estimate a spatial probit model and in the second step (outcome equation) we include an estimated inverse Mills ratio (IMR) as a regressor to control for selection bias. Since the appropriate IMR under SAE depends on a parameter from the second step, both steps are jointly estimated employing the generalized method of moments. The authors explore the finite sample properties of the estimator using simulations and provide an empirical illustration.