In a recent commentary, Kreager (2011) has argued for the “strategic value of studying population composition” in order better to identify the units of population relevant for understanding vulnerability. Conventional approaches, relying on census and survey sources which model populations as homogenous, bounded and composed of discrete households, have failed to accomplish this. Drawing on ethnographic and demographic fieldwork in rural Indonesia, this paper attempts a contribution to ‘compositional demography’ by illustrating a methodology for identifying the subpopulations within a community which are relevant for understanding vulnerability in later life. Key differentials at the local level include the availability of children and socio-economic strata; combining these two dimensions begins to point to significant vulnerable subgroups but falls short of explaining outcomes in later life, because demographic and socio-economic characteristics are mediated by membership of wider networks. The paper therefore characterises a number of indicative and contrasting kinship networks in the study community in terms of their demographic success, marital alliances, land ownership and occupational identities, and examines how these networks have differentially aligned themselves with broader religious, economic and social shifts. Some networks are better able to exploit the opportunities that education or migration entail, others founder on fragmentation of assets, fecklessness and disharmony, and the loss of reputation that entails. It becomes possible to distinguish ascending and descending, locally-bound and translocalizing, and ‘traditional’ and ‘modernising’ networks. Once older people are located within these different kinds of networks, their treatment and differential vulnerability need no longer be modelled purely on their individual characteristics or immediate household or family contexts, but can be understood as the outcome of negotiations, manoeuvrings and sometimes failures of the networks to which they belong. The needs and priorities of elders themselves are rarely of central concern to key agents within the networks, a fact which underlines the limitations of treating ‘the older population’ in isolation. In short, the paper makes an argument for treating networks as the population components of relevance for understanding socio-demographic processes, including population ageing. Identifying and modelling networks raises important conceptual and methodological challenges, but in virtue of being multi-generational, reproductive identities which encompass diversity and cut across strata and space, networks are socially meaningful and empirically grounded population units within which to understand human behaviour.