Pastoral-based animal production systems are under increasing pressure to provide the high quantity and quality of feed needed for optimal ruminant performance. The capacity of farmers to increase forage yield further, solely by increasing fertilizer inputs or through improved pasture management, is limited. Emerging requirements to balance industry production targets against the need to reduce greenhouse gas emissions and N losses pose further challenges. Plant breeding is being asked to deliver results more urgently than at any time previously, and this review attempts to highlight issues that might limit the prospects for future progress by seeking lessons from four past examples: (i) white clover breeding gains and the need to consider the complexity of the grazed grass-clover mixed sward, with its tendency for cycling in plant species composition; (ii) a systems field trial of new and old grass ⁄ clover cultivars, and how the complexity of growth of perennial forage crops, and the dynamic optimality required for sustainable harvesting might limit our ability to breed for 'yield' per se; (iii) the manipulation of a physiological trait (low 'maintenance' respiration) and the implications of such changes for plant fitness and G · E interactions; and (iv) an hypothesis-driven development of a trait (high-sugar grasses) and the value of 'proof of concept' studies, the requirement of scientific understanding of the mechanisms of trait expression, and how one might in future go about assessing breeding achievements. We discuss the general ecological considerations around shifts in the frequency distribution of traits in new populations, whether altered conventionally or by genetic modification, and how selection for a particular trait might inadvertently reduce both fitness and persistence. A major priority for breeding, we propose, might be to revisit previously abandoned traits that affected the physiological performance of forage species, armed now with a capacity to monitor gene expression at the molecular level, and so unravel ⁄ control the G · E interactions that limited their benefits. We also discuss how a 'loss of yield advantage' of new cultivars, seen when tested several years after sowing, requires urgent investigation and propose this might be associated with fitness costs of perenniality. Finally, we argue for a careful reconsideration of what are realistic expectations for systems field trials and that focus on forage breeding might be shifted more to 'proof of concept' studies, critical experimental design, comparing 'traits' rather than 'cultivars', and the wider ecological assessment of fitness and function of traits in the plant, community and ecosystem.