The Loess Plateau in northern Shaanxi has a severely natural environment and is one of the poorest rural areas in China. There is lack of relevantly empirical tests on the sensitivity of poor farmers' livelihood strategies to livelihood capital in different types of villages. Based on survey data of Jia County, this paper divides the sample into three types of villages:sandstorm villages, hilly-gullied villages, and rocky villages. The livelihood capital and livelihood strategies of poor and non-poor households were examined, and the binary logistic regression models were construed to analyze the key factors of shift from purely agricultural to non-agricultural livelihood strategies of poor households. It finds that the increase of human capital, financial capital and social capital had positive effects on the choice of non-farm livelihood strategies of poor farmers, while the increase of natural capital and physical capital had negative effects on the choice of non-farm livelihood strategies. The results for the sandstorm and hilly-gullied villages were similar to the whole sample. While the richer the natural capital was, the more they tended to choose non-farm livelihood strategies in the subsample of the rocky villages. There were significant differences in livelihood strategies among different types of villages, and 70% of poor farmers who did not adjust their livelihood strategies still chose to expand agricultural and forestry production in the future. When making sustainable livelihood strategy adjustments, self-employment and migrant working were their main choices. Natural disasters and the depression of red dates market were the main reasons for their willingness to choose livelihood diversification in the future.