The science of ecology is undergoing many important shifts in emphasis and perspective which have important implications for its role in designing sustainable farming systems. In particular, a shift has occurred from the equilibrium paradigm to one which recognises the dynamic, non-equilibrium nature of ecosystems. Allied to this is the recognition that ecosystems can occur in any one of a number of alternative stable states, depending on the disturbance and management history. An increased emphasis on spatial patchiness in ecosystems has also emerged as appropriate tools have emerged to analyse spatial mosaics. These features have led to a recognition that considerable uncertainty is associated with the outcome of any particular ecosystem modification; hence predictive capacity is also low. Recent considerations of the interrelation between biodiversity and ecosystem function have also explored the questions of how many species need to be in a system to fulfil certain functions and confer resilience. We identify a set of steps that are required for the development of an agricultural system based on mimicking natural ecosystems. Central to this is identifying (1) the functions which are currently suboptimal in the agricultural system, and (2) the species which have key functional roles in the natural system, and then reaching decisions as to the array of species needed to confer system function and resilience.