Abstract. Social skills (e.g., persuading others, showing compassion, staying calm) are of key importance in work and education settings. Accordingly, the goal of many selection processes is to identify candidates who excel in desired skills. For this, high-fidelity simulations such as assessment centers (ACs) are regarded as ideal procedures because they can be used to evoke, observe, and evaluate candidates’ actual behavior. However, research has repeatedly shown that observed performance differences in ACs are not sufficiently driven by the specific skill dimensions that are defined for assessment. Building on multiple speed assessments and incorporating insights from behavioral personality science, we offer an alternative approach for the reliable and valid assessment of distinct social skills. We hereby (a) selected skills on the basis of a bottom-up analysis of observable and distinguishable interpersonal behaviors and (b) specifically designed exercises around these skills (i.e., one skill per exercise, multiple exercises per skill). Here, we present the initial results of this newly developed procedure across three samples in a high-stakes selection context ( N = 589). Generalizability theory analyses showed that a substantial amount of variance in assessor ratings could be attributed to the selected skills. This underlines the importance of more behaviorally focused selection procedures.