Recently, the possibilities for leveraging "big data" in research and pedagogy have given rise to the growing field of "learning analytics" in online education. While much of this work has focused on quantitative metrics, some have called for critical perspectives that interrogate such data as an interplay between technical infrastructures and contingent social practices. Following such calls, this article conceptualizes "learning analytics" as an assemblage of technical, designed, and sociocognitive dimensions. Drawing on DeLanda's articulation of assemblage theory, we examine the ways online learning unfolds within and across these scales by using illustrative quantitative and qualitative data--click-data, user-generated content, and student interviews--from three online higher education courses. We consider how insights generated from such a stance might contribute to critical perspectives on how power circulates in online learning environments--a framework we call "critical learning analytics." We conclude by offering some possibilities for which such a framework might be put to use--not only to map learning analytics as assemblage, but also to imagine how they might be assembled otherwise to promote more ethical instruction and more equitable student flourishing.