Land transfer can promote land scale management, realize economies of scale, and improve scale efficiency. Considering the current level of social and economic development, choosing the most appropriate transfer scale can not only increase households’ income but also improve their capability. Based on survey data from rural households in China’s poverty-stricken areas, this paper uses the threshold regression model to analyze the nonlinear relationship between the land transfer-out scale selection and multidimensional poverty. An endogenous switching regression model, which comprehensively considers the bias and heterogeneity of sample selection, is used to analyze the effect of land transfer on multidimensional poverty. The study yields three main findings. First, the effect of transfer-out rate on the multidimensional poverty index (MPI) structurally changes when the rate reaches 0.7667. Second, based on counterfactual assumptions, we find that high transfer-out land scale can effectively reduce the MPI. Third, if high transfer-out scale households (> 0.7667) choose the low scale, the MPI will increase by 0.0378; if low transfer-out scale households (< 0.7667) have the opportunity to choose the high scale, the MPI will drop by 0.1297. Therefore, to achieve better poverty reduction effects in poor areas, policy maker must rationally guide farmers to transfer sufficient land to cross the threshold value for poverty reduction. The government should formulate different guidance strategies for households to transfer out land to improve their capability and avoid falling into multidimensional poverty. Finally, local governments can alleviate multidimensional poverty by increasing support in poor areas and developing the industrial economy.