In this paper, we study the problem of differential harvest in precision viticulture. Some recent prototypes of grape harvesting machines are supplied with two hoppers and are able to sort two types of grape quality. Given estimated qualities and quantities on the different areas of the vineyard, the problem is to optimize the routing of the grape harvester under several constraints. The main constraints are the amount of first quality grapes to harvest and the capacity of the hoppers. We model the differential harvest problem as a constraint optimization problem. We present preliminary results on real data. We also compare our constraint model to an integer linear programming approach and discuss expressiveness and efficiency.