One of the biggest problems in applications of animal welfare science is our ability to make comparisons between different individuals, both within and across species. Although welfare science provides methods for measuring the welfare of individual animals, there’s no established method for comparing measures between individuals. In this paper I diagnose this problem as one of underdetermination—there are multiple conclusions given the data, arising from two sources of variation that we cannot distinguish: variation in the underlying target variable (welfare experience) (...) and in the relationship of measured indicators to the target. I then describe some of the possible methods of making comparisons, based on the use of similarity assumptions that will have greater or lesser justification in different circumstances, and the alternative methods we may use when direct comparisons are not possible. In the end, all our available options for making welfare comparisons are imperfect, and we need to make explicit context-specific decisions about which will be best for the task at hand while acknowledging their potential limitations. Future developments in our understanding of the biology of sentience will help strengthen our methods of making comparisons. (shrink)