In this paper, we study self-employment in a theoretical setting derived from wage-efficiency spatial models, where leisure and effort at work are complementary. We develop a spatial model of self-employment in which effort at work and commuting are negatively related, and thus the probability of self-employment decreases with "expected" commuting time. We use time-use data from the American Time Use Survey 2003-2014 to analyze the spatial distribution of self-employment across metropolitan areas in the US, focusing on the relationship between commuting time and the probability of self-employment. Our empirical results show that the probability of self-employment is negatively related to the "expected" commuting time, giving empirical support to our theoretical model. Furthermore, we propose a GIS model to show that commuting and self-employment rates are, in relation to unemployment rates, negatively related.