Naturalness is considered important for animals, and is one criterion for assessing how we care for them. However, it is a vague and ambiguous term, which needs definition and assessments suitable for scientific and ethical questions. This paper makes a start on that aim. This paper differentiates the term from other related concepts, such as species-typical behaviour and wellbeing. It identifies contingent ways in which naturalness might be used, as: (i) prompts for further welfare assessment; (ii) a plausible hypothesis for what safeguards wellbeing; (iii) a threshold for what is acceptable; (iv) constraints on what improvements are unacceptable; and (v) demarcating what is not morally wrong, because of a lack of human agency. It then suggests an approach to evaluating animals’ behaviour that is quantitative, is based on reality, and which assesses naturalness by degrees. It proposes classing unaffected wild populations as natural by definition. Where animals might have been affected by humans, they should be compared to the closest population(s) of unaffected animals. This approach could allow us both to assess naturalness scientifically, and to make practical decisions about the behaviour of domestic animals.