The health status of the oldest old, the fastest increasing population segment worldwide, progressively becomes more heterogeneous, and this peculiarity represents a major obstacle to their classification. We compared the effectiveness of four previously proposed criteria (Franceschi et al., 2000; Evert et al., 2003; Condo et al., 2006; Andersen-Ranberg et al., 2001) in 1160 phenotypically fully characterized Italian siblings of 90 years of age and older (90+, mean age: 93 years; age range: 90-106 years) belonging to 552 sib-ships, recruited in Northern, Central and Southern Italy within the EU-funded project GEHA, followed for a six-year-survival. Main findings were: (i) "healthy" subjects varied within a large range, i.e. 5.2% (Condo), 8.7% (Evert), 17.7% (Franceschi), and 28.5% (Andersen-Ranberg); (ii) Central Italy subjects showed better health than those from Northern and Southern Italy; (iii) mortality risk was correlated with health status independently of geographical areas; and (iv) 90+ males, although fewer in number, were healthier than females, but with no survival advantage. In conclusion, we identified a modified version of Andersen-Ranberg criteria, based on the concomitant assessment of two basic domains (cognitive, SMMSE; physical, ADL), called "Simple Model of Functional Status" (SMFS), as the most effective proxy to distinguish healthy from not-healthy subjects. This model showed that health status was correlated within sib-ships, suggesting a familial/genetic component. (C) 2013 Elsevier Ireland Ltd. All rights reserved.