The paradigmatic methods of scientific investigation have been the control-experimental or “intervention” methods. Although not without their epistemic concerns, both in principle and in practice, the methods have nonetheless been thought to constitute the ideal of scientific investigation. Certain recent data-science practices employ methods that resemble and differ from these methods in a number of important ways. In this article, my concern will be with some of the ways in which the methods in these practices differ from control-experimental methods and with the epistemic implications these might be thought to have. I argue that, for most points of epistemic difference between the methods, any implied epistemic difference is superficial and that data-scientific methods can be understood to involve an implicit epistemic relation identical to that of control-experimental methods. Towards the end of the article, I examine a point of difference between data-scientific and control-experimental methods that might, in a subset of cases, be thought to have genuine epistemic significance. This point of difference is in the machine-learning techniques employed in the former, some of which techniques have implications, not for the mentioned epistemic relation as such, but for the possible content of the claim established by the investigation.