Analytic systems increasingly allow companies to draw inferences about users’ characteristics, yet users may not fully understand these systems due to their complex and often unintuitive nature. In this paper, we investigate inference literacy: the beliefs and misconceptions people have about how companies collect and make inferences from their data. We interviewed 21 non-student participants with a high school education, finding that few believed companies can make the type of deeply personal inferences that companies now routinely make through machine learning. Instead, most participant’s inference literacy beliefs clustered around one of two main concepts: one cluster believed companies make inferences about a person based largely on a priori stereotyping, using directly gathered demographic data; the other cluster believed that companies make inferences based on computer processing of online behavioral data, but often expected these inferences to be limited to straightforward intuitions. We also find evidence that cultural models related to income and ethnicity influence the assumptions that users make about their own role in the data economy. We share implications for research, design, and policy on tech savviness, digital inequality, and potential inference literacy interventions.