One of the most important applications of precision agriculture to viticulture is the management of the quality of grapes at the field level through selective harvest. The recent developments of prototypes of conventional grape harvesting machines are able to sort two types of harvest quality. In this paper, we propose a formal description of the problem of differential harvesting. The objective is to optimize the routing of the grape harvester under several constraints. Our approach is based on Constraint Programming (CP), which is a powerful paradigm to solve combinatorial problems. We give a formulation of the problem as a Constraint Optimization Problem. The last section of the paper is devoted to preliminary experimental results on real data from a French vineyard.