In this paper, we develop a theoretical model of land leasing that includes transaction costs, risk pooling motives and non-tradable productive inputs, and investigate the empirical implications of land contracts using data collected from four villages in Ethiopia. We show that sharecropping is the dominant contract if transaction costs are negligible, but that a rental contract may arise if transaction costs decrease with increasing the tenant s share of output. When this is the case, the theory predicts that area operated by tenants will be an increasing function of their land endowment and that fixed rental contracts will be more likely in situations where transaction costs are higher. We find empirical support for these predictions in the villages studied. We also find that input of labor per hectare is about 25% lower on sharecropped than on other land tenure types, but that the differences in total value of inputs, outputs and profits per hectare are statistically insignificant and relatively small in magnitude. These results support the Marshallian argument that sharecropping reduces labor effort, but also support the New School perspective since the magnitude of the inefficiency is relatively small. A bigger source of inefficiency (and inequity) in the study villages appears to be the limited lease market for oxen services, together with credit constraints that limit the ability of land and oxen poor households to purchase oxen.