Why some countries are so much richer than others is a question of central interest in economics. Low aggregate income per worker in poor countries is mostly accounted for by low labor productivity and high employment in agriculture. This thesis attempts to understand cross-country income difference through examining productivity differences at the sector level - in agriculture and in non-agriculture. Between rich and poor countries, there is a 45-fold difference in agricultural output per worker and a 34-fold difference in mean farm size. In the first chapter, I argue farmer's skill as a plausible explanation for these differences. The model features heterogeneity in innate agricultural skill, on-the-job skill accumulation, and span-of-control in agricultural production. I show that low total factor productivity (TFP) in poor countries not only induces more individuals with low innate skill to choose farming, but also reduces the incentive to accumulate skill. Between rich and poor countries, the model generates substantial difference in farmer's skill, which translates into differences in agricultural productivity and farm size distribution. Quantitatively, the calibrated model explains half of the cross-country differences in agricultural output per worker, and successfully replicates the size distribution of farms in both rich and poor countries. Cross-country productivity differences are asymmetric across sectors. The labor productivity gap between rich and poor countries in agriculture is twice as large as that in the aggregate, and ten times larger than that in non-agriculture. The second chapter shows that these sectoral productivity differences can arise solely from difference in aggregate TFP. I extend the framework in the first chapter to allow for different skill in non-agricultural production as well. Low TFP distorts the allocation of skills across sectors and discourages skill accumulation on the job. To discipline the initial skill distribution and skill accumulation, the model is calibrated to match earnings distribution and age-earnings profiles in both agriculture and non-agriculture in the U.S. The model's implications are then examined using a sample of 70 countries that covers a wide range of development. Between rich and poor countries, the model accounts for most of the productivity differences at the sector level - productivity difference in agriculture in the model is 1.8 times larger than those in the aggregate and 6 times larger than those in non-agriculture. As in the data, the share of farmer in the labor force in the model declines from 85 percent in the poorest countries to less than 2 percent in the richest countries. These results suggest that policy aiming at improving overall efficiency should be prioritized.