This thesis examines how the agency of working children relates to the nature of their work and the harm caused by it. Theorists and practitioners specialised in children's work have argued that its harms should be understood from the perspectives of working children and that efforts to improve their situation should involve them and meet the interests they express. Their approach is premised on children's capacity for autonomous and rational decision making. The thesis develops an alternative approach, by examining harm in children's work and children's responses to it with an understanding of agency as being conditioned by material and social contexts.Its theoretical purpose is to use Bourdieu's theory to examine children's work. Its methodological contribution is that it studies children's work as a practice, rather than children's individual experiences and perspectives on their work. This involved investigation of patterns characterising forms of children's work, and exploration of why these patterns exist and how they might be changing which focuses on how children are involved and affected. The thesis is based on empirical study of children's work in cement block construction in peri-urban localities, as apprentices in Calavi, Benin, and as unskilled workers in northern Bengaluru, in the state of Karnataka, India. Construction is recognised as a worst form of children's work by the ILO, but the work studied was locally condoned. In Calavi, apprenticeship was considered as professional training, and in Bengaluru, children's construction work contributed to family livelihoods. These are the kind of work situations that social scientists who stress children's agency have suggested are likely to be beneficial. Main sources of data were observations of construction work and interviews with workers, mostly children, as well as their direct employers. Interviewed children did not see their work as seriously harmful, although it was found to risk impairing their physical integrity and to confirm their inequality. In Calavi, children were much more oppressed in their work than children in Bengaluru, but in both sites children acted with reasons and interests. They did not however act to change harmful work conditions. Analysis shows how their age, gender and class positions might have shaped their perspectives in ways which explain why they largely accepted them. The children's shared hope that their own children would not work as they had indicated their involvement in social change which might be undermining their work practices. The findings confirm the importance of examining children's perspectives in attempt to understand the causes and consequences of their work. Yet they suggest that children may not always be able to identify harm, and thus the relevance of pursued efforts to develop ways of studying harm in children's work which do not assume their capacity for autonomous and rational decision making or rely primarily on their perspectives.