Forest plays an important role in keeping water ecosystem services, such as drinking water provision. Thus, payment for ecosystem services is an essential instrument to promote forest restoration in agricultural watersheds. However, funds are limited and must be well planned to ensure water resources conservation and water ecosystem services improvement. In this context, our study aimed to identify priority areas for forest restoration, based on water ecosystem services in agricultural landscapes. For this, we have developed a decision-making support model for agricultural watersheds (in the Atlantic Forest region), based on mixed approaches, that were multicriteria evaluation (MCE) and Participatory Technique. The model will help decision-makers and stakeholders to set priorities for payment for ecosystem services programs implementation. So, we evaluate its application in watersheds with different forest cover patterns to check if it can be applied to different landscape patterns. The base of the model was the following criteria, that were produced with high-resolution data and ranking in the Participatory Technique context, considering their importance for the study: proximity to spring, slope, soil erodibility, topographic index, and land-use/land-cover (LULC). The criteria were aggregated by the Weighted Linear Combination (WLC) method (an MCE method). The priorities maps showed areas classified as high priority near the rivers (at most 200 m far from rivers), on the greatest slopes (>40%), on soils associated with high potential of erosion, and predominantly in agriculture lands. However, this class presented more percentage of the area associated with native forest in the forested watershed (native forest covers 55% of its area) than in the watershed non-forested (native forest covers 25%). Another important point of the final maps was a high percentage of areas associated with the medium class, which is a characteristic of the WLC method. Thus, areas classified as high and medium priority was defined as targets for forest restoration in the watersheds. We can conclude that for small watersheds, the MCE method, with high-resolution data, supports an appropriate prioritization of areas for forest restoration, aiming at the improvement of water ecosystem services. This way, our model can be applied to various payments for ecosystem services schemes in agricultural landscapes worldwide.