Abstract:Traditional theories in agricultural economics generally assume there is a constant return to scale in agriculture.Such an assumption is not only inconsistent with the concept of the endogenous technical change and the “learning by doing” models in neoclassical economics,but also can’t explain the agricultural growth caused by the intensive and meticulous farming practices in traditional peasant economy and the division of labor in modern agriculture.By expanding the “learning by doing” models developed by Arrow and Yang,this paper analyzes how farmers can achieve the increasing return to scale in the process of “learning by doing” by adjusting the planting structure and agricultural inputs in a traditional peasant economy and an era of agricultural division of labor without external technical change.The results indicate that (1) without rigid constraint in agricultural labor,the “learning by doing” will improve labor productivity and increase return to scale when the farmers rent more farmland or increase the frequency of crop planting to make full use of agricultural labor force; (2) as agricultural labor force transfer to non-agricultural acceleratedly,the return to scale in agriculture become decreasing after the increase induced by the learning by doing due to the rigid constraint of agricultural labor force; (3) in order to overcome the rigid constraints of agricultural labor force,farmers will spontaneously participate in the agricultural division of labor,so as to speed up the agricultural "learning by doing" and improve the degree of agricultural specialization,which induce the increase of agricultural returns to scale.The findings show that agriculture has never been a passive industry with low efficiency,but evolving industry with constant pursuit of efficiency and growth.