Animal welfare is generally referred to as the quality of an animal's life as experienced by the individual animal. On-farm welfare assessment, however, usually relies on both individual and group measures. As regards the latter, individual animals are not identified (eg incidence of stereotypic behaviour in a pen) or features of the whole group (eg score obtained from qualitative behaviour assessment) are used. This raises the question whether our current approaches to on-farm assessment sufficiently consider the individual nature of animal welfare. Measures assessed at the group level bear the disadvantage that distribution across group members may be skewed and the most affected individuals are not necessarily identified. However, the importance of knowing about the welfare state of individual animals depends on the purpose of the assessment. If the primary aim is farm assurance, the individual animal is of lesser importance, but non-compliance with thresholds at herd/farm level or comparison with peer farms may induce change. Using individual measures in a sample of animals means that animals not sampled but requiring intervention, eg for treatment of lameness, would have to be identified subsequently. Measures truly taken at the group level make individual interventions difficult, but interventions implemented at the group level (eg reducing stocking density) do not necessarily require information on the individual animal. Automatic detection of welfare-relevant states has received increased attention and identifying individual animals with impaired welfare seems to be promising. Automated early detection of problems may also reduce the ethical dilemma that traditional assessments at the end of the production cycle, eg in broiler chickens, may identify welfare impairments but not directly benefit the affected animals. Reflection on individual and group measures and their consequences for animal welfare may help in interpreting the outcomes of the assessments and stimulate future developments in the field.