Studies have shown that individual differences in word reading can be observed for both skilled and novice readers. Several factors that could cause individual differences including reading experience, reading capacity, and oral language have been investigated. However, little is known about the influence of reading instruction on individual differences in reading. Given that early reading, training is critical to help children become proficient readers, the influence of reading instruction on subsequent reading behaviours should also be well understood. Thus, in this study, we investigated the relationships between reading instruction and individual differences in reading using computational models of reading. The model was exposed to a sound-focused, meaning-focused or balanced training scheme. We quantified the model’s reliance on accessing semantics for reading, as an index of individual differences in semantic reliance (SR). The simulation results demonstrated that the degree of SR depended on reading instruction. Meaning-focused training resulted in higher SR, and that was followed by balanced training and then sound-focused. Moreover, SR was able to predict the model’s word reading performance and interacted with other psycholinguistic reading factors including frequency, consistency, and orthographic neighbourhood size.