Malnutrition occurs frequently in the elderly with important clinical and functional consequences. Moreover, the treatment of malnutrition in the elderly may be effective if clinical and nutritional interventions are performed in the early stages. Therefore the early identification of the risk of malnutrition using validated and handy tools plays a pivotal role in terms of clinical outcome. Mini Nutritional Assessment (MNA) was validated for this purpose since many years but it is still ongoing the debate over whether the use of different items in certain clinical conditions can be effective without affecting the validity of the nutritional status evaluation. The aim of this study was to assess the agreement between different versions of MNA in the evaluation of nutritional risk in elderly subjects. Methods: 522 subjects, 345 women and 177 men, were recruited from nursing homes or were free living in three different regions in Italy. All subjects underwent a multidimensional geriatric evaluation, addressed especially to nutritional status. We compared three different versions of MNA: the "original" version; a "proportional" MNA (MNA-P) in which the total MNA score was replaced by the ratio between the maximum score that each subject could obtain without including the body mass index (BMI) and the total original MNA score; and a third version in which calf circumference (CC) and mid-upper arm circumference (MAC) were used instead of BMI. Results: According to the original MNA, a high prevalence of malnutrition was found out in both genders (26% of women and 16.3% of men); both the versions of MNA, in which BMI was not considered, showed a good predictive value compared to original MNA. In particular, the MNA-P. showed an overall efficiency equal to 89,1% with specificity and positive predictive value respectively equal to 97.5% and 95.2%. MNA-CC-MAC showed even better results in terms of overall efficiency (91.4%), sensitivity (81.1%), specificity (97.1%), positive and negative predictive values (94.2% and 94.4%, respectively). Conclusion: The different versions of MNA gave similar results in the classifications of subjects and in comparison with nutritional and biochemical parameters. Moreover MNA versions that did not considered BMI seem to be more effective in singling out subjects with risk factors related to malnutrition (disability, reduced strength and calf circumference, anaemia). © 2013 Serdi and Springer-Verlag France.