We shed light on poverty dynamics by analyzing a census-like survey of three villages of Guizhou province during 2004 to 2011. While the absolute poverty rate is decreasing sharply in the sample, households are highly vulnerable to shocks, and rates of entry or re-entry into poverty are high. We decompose measures of wealth to reveal the proximate causes of poverty entry or exit, then use logistic regression and multivariate hazard models to look for underlying causes. Agricultural income contributes most to changes in poverty status. Poverty entry and exit are both related to household characteristics, assets, and social capital. Rural-urban migration strongly increases the probability of poverty exit, as do government transfers. The frequent changes in poverty status highlight the importance of improving policy targeting and/or implementing village-wide poverty-alleviation strategies.