How society responds to rising human food needs will be pivotal to the future of global biodiversity1–3. Cutting food waste and consumption of ruminant meat are essential2–5 . On the supply side, detailed field data from five continents consistently show extinctions would be greatly reduced if residual demand could be met by land sparing - boosting yields (production per unit area) on existing farmland while conserving (or restoring) remaining natural habitats6,7. But limiting the land cost of agriculture through high-yield farming raises other important concerns because when expressed per unit area, high-yield systems can generate high levels of negative externalities such as greenhouse gas (GHG) emissions and nutrient emissions 8,9losses8,9. However, such metrics systematically underestimate the overall impacts of lower-yield systems. Here we instead develop a framework quantifying externality costs (including off-site effects) per unit of production. Applying this more comprehensive approach to five key externalities and four major farm sectors, we discover that, rather than associations between externality and land costs of alternative production systems being characterised by trade-offs (i.e. negative correlations), those for which we could find data more commonly exhibit positive correlations. Per- unit production, systems which take up less land often generate fewer externalities. For GHG emissions (the best-documented externality) these associations become more strongly positive once the effects of land use are included. Our conclusions are limited by data availability – remarkably few studies quantify externalities alongside yields, and many important externalities remain largely unmeasured. Moreover, some high-yield systems we examined have high externality costs, even per- unit of production, and none can generate environmental benefits unless linked with efforts to limit rebound effects10,11. However, our results identify several promising ways of increasing yields while lowering other environmental impacts, and more generally suggest that trade-offs among key cost metrics are not as ubiquitous as sometimes perceived.