The appeal of non-probabilistic surveys has been on the rise given the costs and decreasing response rates associated with probabilistic surveys. Yet the non-random selection of respondents into non-probabilistic surveys leads to inaccurate estimates if there are systematic differences in relation to the variable of interest between the self-selected respondents to the survey and the rest of the target population. In addition, for non probability samples there is no general statistical theory that justifies when and why accurate inferences can be expected. This paper presents a review of established uses of non-probability samples in comparative, cross-national contexts and their value for policy. In particular, the review focuses on the rationales for using non-probability samples, the risks involved and the potential ways of mitigating these risks. The paper concludes by providing some potential roles non-probability samples could play in the context of the OECD Teaching and Learning International Survey (TALIS).