Yohe and Tol (2002. Global Environmental Change 12, 25-40) built an indexing method for vulnerability based on the hypothesis that the adaptive capacity for any system facing a vector of external stresses could be explained by the weakest of its underlying determinants-the so-called "weakest link" hypothesis. Their structure noted eight determinants, but the approach could handle any number. They quoted analogies in support of the hypothesis, but loose inference is hardly sufficient to confirm such a claim. We respond to this omission by offering an empirical investigation of its validity. We estimate a structural form designed to accommodate the full range of possible interactions across sets of underlying determinants. The perfect complement case of the pure "weakest-link" formulation lies on one extreme, and the perfect substitute case where each determinant can compensate for all others at constant rates is the other limiting case. For vulnerability to natural disasters, infant mortality and drinking water treatment, we find qualified support for a modified weakest link hypothesis: the weakest indicator plays an important role because other factors can compensate (with increasing difficulty). For life expectancy, sanitation and nutrition, we find a relationship that is close to linear-the perfect substitute case where the various determinants of adaptive capacity can compensate for each other with relative and persistent ease. Moreover, since the factors from which systems derive their adaptive capacities are different for different risks, we have identified another source of diversity in the assessment of vulnerability. © 2006 Elsevier Ltd. All rights reserved.