How do children learn about the structure of the social world? We tested whether children would extract patterns from an agent's social choices to make inferences about multiple groups' relative social standing. In Experiment 1, 4- to 6-year-old children (N = 36; tested in Central New York) saw an agent and three groups ("Group-A," "Group-B," and "Group-C") and observed the agent choose between pairs of individuals from different groups. Across pairwise selections, a pattern emerged: The agent chose individuals from "Group-A" > "Group-B" > "Group-C." Children tracked the agent's choices to predict that "Group-A" was "most-preferred" and the "leader" and that "Group-C" was "least-preferred" and the "helper." In Experiments 2 and 3, we examined children's reasoning about a more complex pattern involving four groups and tested a wider age range. In Experiment 2, 5- to 10-year-old children (N = 98; tested in Central New York) used the agent's pattern of pairwise choices to infer that the agent liked "Group-A" > "Group-B" > "Group-C" > "Group-D" and to make predictions about which groups were likely to be "leaders" and "helpers." In Experiment 3, we found evidence for social specificity in children's reasoning: 5- to 10-year-old children (N = 96; from 26 US States) made inferences about groups' relative "social" but not "physical" power from the agent's pattern of affiliative choices across the four groups. These findings showcase a mechanism through which children may learn about societal-level hierarchies through the patterns they observe over time in people's group-based social choices.