To understand the persistent social class achievement gap, researchers have investigated how educational settings affect lower vs. higher socio-economic status (SES) students’ performance. We move beyond the question of actual performance to study its assessment by evaluators. We hypothesized that even in the absence of performance differences, assessment’s function of selection (i.e., compare, rank and track students) leads evaluators to create a SES achievement gap. In two experiments (N = 196; N = 259), participants had to assess a test supposedly produced by a high- or a low-SES student, and used assessment for selection (i.e. normative grading) or learning (i.e. formative comments). Results showed that evaluators using assessment for selection found more mistakes if the test was attributed to a low- rather than a high-SES student, a difference reduced in the assessment for learning condition. The third and fourth experiments (N = 374; N = 306) directly manipulated the function of assessment to investigate whether the production of the social class achievement gap was facilitated by the function of selection to a greater extent than the educational function. Results of Experiment 3 supported this hypothesis. The effect did not reach significance for Experiment 4, but an internal meta-analysis confirmed that assessment used for selection led evaluators to create a SES achievement gap more than assessment used for learning, thereby contributing to the reproduction of social inequalities.