The delimitation of within-field zones has become one compulsory step in most of the processing chains of precision agriculture data. This study proposes a global optimization approach using constraint programming to improve the delineation of within-field zones. The optimization process involved two major constraints: (1) the final zones must have an area larger than an expert threshold to cope with operational constraints; (2) adjacent zones must have a significant difference in attribute value so that the resulting zones are meaningful for the decision-making process. The proposed constrained programming approach performs the optimal merging of zones that minimizes the sum of squared differences between each observation's attribute and the mean zone attribute in which the observation belongs, while fulfilling all of the aforementioned constraints.