Rule-governed behavior (RGB) results from contact with a verbal description of a contingency as opposed to prior contact with that contingency. Despite its importance, research on the establishment of RGB with learners who do not display the skill is limited. Tarbox, Zuckerman, Bishop, Olive, and O’Hora (The Analysis of Verbal Behavior, 27, 125-139, 2011) used multiple-exemplar training (MET) to teach children with autism spectrum disorder to follow rules specifying an antecedent and a behavior. We conducted a systematic replication of the Tarbox et al. study with three boys diagnosed with autism spectrum disorder and extended those methods to rules specifying a behavior and either a preferred or nonpreferred consequence (e.g., “If you clap, then you get candy”). In baseline, participants typically followed a given instruction regardless of whether the consequence was preferred or nonpreferred. Following MET, all participants responded accurately to novel rules, indicating that MET may be an effective method to establish basic RGB repertoires.