The human auditory system is thought to represent regularities inherent in auditory information in internal predictive models. Sounds not conforming to an established regularity (deviants) elicit prediction error signals in the brain, alerting the auditory system to information not explainable based on currently active models. Here, we examine the widely neglected characteristic of deviants bearing predictive information themselves. In a modified version of the oddball paradigm using higher-order regularities, we set up different expectations regarding the next sound following a deviant. In a deviant detection task, participants listened to an oddball sequence including two deviant types following diametrically opposed rules: one deviant occurred mostly in succession (high repetition probability) and the other mostly in isolation (low repetition probability). To probe whether explicit knowledge facilitates rule extraction, participants in Experiment 1 were naïve, whereas in Experiment 2 they were made aware of the rules. In both experiments response times significantly decreased from first to second deviant when repetition probability was high – albeit more in the presence of explicit rule knowledge. There was no evidence of a facilitation effect when repetition probability was low. Significantly more false alarms occurred in response to standards following high compared to low repetition probability deviants, but only when participants were aware of the repetition rules. These findings provide evidence that deviants can be integrated into predictive models, enabling predictions about auditory events in the imminent future. More generally, this new paradigm may provide further insights into the predictive properties of the human brain.