The assessment of animal welfare is a complex subject which gives rise to divergent views and debate. It is generally accepted that scientific welfare assessment must involve multidisciplinary approaches, and that to interpret results unambiguously, a high level of control over the experimental conditions is required. Such considerations would appear to militate against attempts to measure welfare in a practical farm situation, where systems are relatively uncontrolled and contain many confounding factors to complicate interpretation. In consequence, fundamental welfare scientists sometimes consider that on-farm welfare assessment is of limited value. However, adherents emphasise that on-farm application is the final objective of all livestock welfare science endeavours, and also gives unique options for large-scale population studies and access to a diversity of environmental circumstances. On-farm welfare assessment not only provides an opportunity for extending knowledge on animal requirements, but is also a necessary tool in the growing requirement to assess and certify animal welfare status for legislators and consumers. However, the economic and time limitations, combined with difficulty of close access to individual animals, restrict the range and detail of possible measures. It is also essential that a consensus exists that the measurements taken are objective and meaningful to stakeholders. These constraints have tended to drive the techniques used in Farm Assurance schemes towards assessment of resource provision and management records. However, animal-based measures of health and behaviour are now being more widely explored, and the validation and standardisation of simple integrative measures for such approaches is an important future development.