One of the main questions regarding the phenomenon of part-time farming is whether it is a stable situation or a temporary stage before quitting farming altogether. A conclusion may be reached by examining the structural state dependence of farmers' discrete choices regarding off-farm work over time. A dynamic programming model of utility maximization in which current work decisions affect future utility is developed, and an estimation strategy based on the Hotz-Miller Multinomial Logit framework for panel data is suggested. The model is estimated using Israeli census-panel data and the hypothesis that state dependence is not important is strongly rejected. In particular, the results indicate that farmers choose to work off the farm if it leads to lower prospects of a future exit. Hence, it may be concluded that Israeli farmers do not see the off-farm work option as a step on the way out of agriculture, but rather as a preferred long-run choice in combination with farming.