OBJECTIVE: To assess the cost of malnutrition and related comorbidities among elderly patients living in the community and determine the value of nutritional support. METHODS: Observational, prospective, longitudinal, cohort study with a 12 months follow-up, with 90 general practitioners in France. Two groups of doctors were selected based on historical prescribing practice: one with frequent (FNS) and one with limited (LNS) use of nutritional support. Three hundred seventy-eight elderly patients aged over 70 living in the community, either at home or in institutions, at risk of malnutrition or with established malnutrition. The observational design was preferred over the randomized option because it would not have been ethical to ask doctors to refrain from providing nutritional support when this was their normal practice. To adjust for baseline differences between groups, costs were analyzed using a propensity score model. Confidence intervals were estimated using bootstrap methods. RESULTS: Population in the two groups were balanced for age, gender, weight and body mass index. The groups differed significantly in terms of housing status (p<0.05) and nutritional status (p<0.001). Adjusted costs per patient of hospital care (-€551), nursing care (-€145) and other medical care were significantly reduced in the FNS group as compared to the LNS group, with cost savings of -€723 (90% CI: -€1.444 to -€43). Including oral supplementation costs, the total cost savings per patient attributable to nutrition support were -€195 (90% CI: -€929 to +€478). CONCLUSION: Appropriate nutrition diagnosis and support may contribute to reduce the costs of health care. Propensity score models are a valuable framework for the analysis of cost data, when it is not possible to conduct randomized studies.