Adults use an individual's behavior in one moral subdomain to make inferences about how they will act in another moral subdomain, reflecting a tendency to attribute underlying traits to individuals. We recruited 4- to 7-year-old children from a large city in North America to investigate their ability to generalize from one moral subdomain to another and integrate these pieces of information to form trust and friendship decisions, focusing on the subdomains of helping and fairness, given their centrality to moral cognition. In Experiment 1 (N = 131; 49% female; 38% White), children watched a protagonist help or hinder another person with their goal and then engage in either a fair or an unfair resource distribution between two novel recipients; in Experiment 2 (N = 130; 52% female; 55% White), these events were reversed. We recorded the children's surprise at the second event and their willingness to trust subsequent information provided by the protagonist and to befriend her. Children selectively generalized from the initial behavior, reporting greater surprise to fair (vs. unfair) behavior after the protagonist hindered and greater surprise to the protagonist helping (vs. hindering) after she distributed resources unfairly previously. Moreover, the presence of a single moral transgression lowered children's trust and friendship judgments to chance levels. These findings demonstrate that moral transgressions (vs. moral adherence to moral norms) provide a basis for guiding children's subsequent expectations for future behavior across moral subdomains, as well as for forming social decisions regarding whether to befriend and trust individuals, for children as young as age 4 years.