High-variability (HV) training is a technique that uses a wide variety of tokens representative of a specific category’s internal variation. Readers should be able to handle material written in both easy-to-read and stylized formats. The conventional approach to reading instruction is to start with low-variable examples before spreading out to less common cases. This study performs two experiments comparing low-variability (single source) training against HV (multiple source) training of novice readers recognizing Chinese characters. Participants were tested on their ability to recognize characters they had trained with as well as their ability to generalize their training to unfamiliar variations. Results showed that HV training was associated with decreases in response time and increases in overall accuracy, including the ability to generalize training to novel variation not seen during training. This study found evidence showing that HV training is an effective method for English speakers learning to recognize Chinese characters.