Most empirical studies of the child labor versus schooling choice ignore a common third option in Africa: to send a child away from the household to work elsewhere. The reason for this shortcoming is that most datasets are based on household surveys which only provide information on members living in the household at the time of the interview. An important part of the child labor choice, the decision to send a child away to work, is therefore generally ignored in ordinary survey-based studies. This thesis adds to the existing child labor literature by focusing on the child labor migration choice. The dataset used contains information about all children born from the same mother and include their current location and schooling status, as well as the motive they had for leaving. Child labor migration is an important social policy issue in West Africa. The practice is common throughout the region and the children involved tend to be even more vulnerable than other child workers because they leave the social safety nets of family and kin behind. In many West African countries, the phenomenon is indiscriminately labeled as child trafficking, but this may be a misrepresentationof the situation of many supposedly trafficked children, as the motivation, dynamics and conditions may vary greatly from one case to another. As a result, numerous anti-trafficking projects have been based on poorly documented and unwarranted assumptions about the determinants of the practice, and they have until now been largely ineffective.The model applied to analyze the determinants of child labor migration in this thesis is based on three hypotheses for explaining child labor, as proposed by Bahlotra and Tzannatos (2003): weak incentives to schooling relative to work options, (binding) poverty constraints, and limitations to parental (or agent) altruism. The three potential explanations have different policy implications. Thefirst would call for improving incentives to education (education policy), the second suggests social policy interventions, while the third would give legitimacy to legal measures such as sanctions. What would be an adequate mix of these policies? By applying a multinomial logit model to data from Benin, indicators for the three hypotheses are tested while controlling for cultural and demographic factors. The results of the analysis only partially support the theory. Incentive structures to local labor and schooling turn out to function more or less as expected relative to the schooling choice, but not to the child labor migration choice. While poverty has the expected impact on increasing the labor migration of girls, this is not the case for boys. Credit availability was expected to relaxpoverty constraints, but instead appears to have other functions vis-à-vis the child labor migration choice. Features of the child’s agents matter but not systematically in accordance with the theory. The fact that better educated mothers have a higher propensity to send their daughters away to work and better educated household heads a higher propensity to send their sons do raise some questions as to the possible limitations to parental altruism, since ignorance intheir case is harder to blame. But it also raises the question of whether (all) child labor migration is as bad as assumed, at least relative to the options the child has if staying at home. The mere complexity of the results should serve as a warning against jumping to intuitively appealing but not scientifically supported conclusions on how the problem could be addressed. In this sense, it would be advisable to bring up some of the issues raised by the research results in a dialogue with communities at risk. The thesis concludes by offering suggestions on how to define furtherresearch in a way that helps clarify some of the core questions raised by the study. Based on the lessons from the local dialogue and the results from a second survey, a more adequate policy to prevent the harmful aspects of child labor migration can be defined in the future.