This paper uses unique primary data on directly elicited individual subjective expectations to analyse and characterize the process that generates the income of poor, rural Indian households. We validate and use responses to subjective expectations questions and a parametric assumption to fit a household-specific probability distribution for future income. Combining computed moments from this distribution with data for actual current income, we specify and estimate a dynamic model of household income. We find that our households face a very persistent income process. Our paper is one of the first that uses subjective expectations data to model income processes.