Farm labor supply of farm operators is important for the analysis of their off-farm labor supply. We use a unique data set which includes such information to demonstrate its importance. Other studies had to use implicit assumptions in order to proxy the marginal product of farm labor with observable farm attributes. We find that these assumptions are too strong. We do it by estimating the off-farm participation equation over different subsamples defined according to the level of farm work. We correct for selection by estimating an endogenous switching regression model of the off-farm participation decision, in which the selection criterion is farm participation. Selection is found to be significant for farm workers only. The qualitative conclusions are unaffected by controlling for selection.