To make inferences about the frequency of events, people often exploit observations of relevant instances sampled from their personal social network. We examined how this ability develops ontogenetically and conducted a study in which children and adults judged the relative frequencies of first names in Germany. Based on the recalled instances of the names in participants’ social networks, we modeled their frequency judgments and the underlying search process with a Bayesian hierarchical latent-mixture approach. We found that the judgments of most adults were best described by a noncompensatory strategy that assumes limited and sequentially ordered search, whereas the judgments of most children were best described by a compensatory strategy that assumes exhaustive search and information aggregation. These results highlight that already children use instance knowledge to infer event frequencies and suggest that the ability to conduct ordered and focused search is central to the development of noncompensatory instance-based inference.