Broadly speaking, ecology seeks to understand the structure and dynamics of individual populations of plants and animals, of communities of interacting populations, and of ecosystems. Ideally--the reductionist dream--it would be nice to build such an understanding of how individual populations respond to disturbance upon a fundamental understanding of the behaviour and physiology of the constituent individuals. This is vastly ambitious. In practice, most successful applications of population biology, for instance to the management of harvested systems or to the control of pests and pathogens, have treated the population itself as the basic variable (in equations whose parameters are assessed phenomenologically, even though they are in principal derivable from the more basic parameters pertaining to the behavioural ecology of individuals). By the same token, studies of the structure and function of communities and ecosystems are often and usefully approached phenomenologically as things in themselves (topology of food webs, 'plumbing diagrams' of energy flows, etc.) rather than being derived in a more fundamental way from detailed consideration of the interaction among the constituent populations. Furthermore, at every level of approach, it is often difficult to perform meaningful experiments which control variables and isolate single factors in a tidy way. Frequently, the spatial or temporal scale is such that observational data and/or past records provide the only way to estimate parameters. In the 1970s and early 1980s this inability of much of ecology to conform rigidly to simplistic schemes of 'how science is done' caused much angst. In this paper I offer a grandly opinionated overview of these issues.