In our study, we tested the hypothesis that feature-based and rule-based generalization involve different types of processes that may affect each other producing different results depending on time constraints and on how generalization is measured. For this purpose, participants in our experiments learned cue–outcome relationships that followed the opposites rule: Single cues that signaled the same outcome (e.g., A-1/B-1) predicted the opposite outcome when presented in compound (e.g., AB-2). Some cues were only presented in compound during training (e.g., EF-1) to see if at test participants tended to generalize according to rule-based (i.e., E-2/F-2) or according to feature-based generalization (i.e., E-1/F-1). The generalization test used 2 different tasks: a predictive judgment task, and a cued-response priming task. In Experiment 1, participants’ verbal ratings were consistent with rule-based generalization. However, participants’ reaction times (RTs) in the cued-response priming task were consistent with feature-based generalization. Experiment 2 replicated the results from Experiment 1, and it also provided evidence consistent with feature-based or rule-based generalization depending on whether a short stimulus onset asynchrony (SOA; 200 ms) or a long SOA (1300 ms), respectively, was used in the priming task. Our results are interpreted as supporting the idea that feature-based generalization process relies on fast, associative processes, whereas rule-based generalization is slow and depends on executive control resources. The latter generalization process would inhibit the former when enough time and resources are available. Otherwise, feature-based generalization would take control of responses.