Canonical ideologies tend to position datasets as neutral representational tools, when datasets may be more aptly characterized as power-laden systems for signification. While critical for interpreting the cultural meaning of data, the skills needed to historicize, situate, and deconstruct datasets are often underrepresented in STEM education. In this talk, I outline a series of pedagogical approaches to teaching cultural analysis of datasets. I show how, by cultivating competency in hermeneutics, ethnography, and critical theory, students can learn to attend to the cultural provenance of datasets across a number of registers - from interrogating the belief systems of data designers, to examining the cultural logics of data infrastructures, to analyzing the interests of data-producing institutions, to unpacking the discourses that shape public understandings of data. Further, by pluralizing the epistemic lenses through which data are analyzed, students have an opportunity to nourish reflexive sensibilities - discerning their own cultural positioning as they question why culture tends to be deleted from data science work.