The body of work presented here explores the assumption of uniformity in skilled reading which assumes that all skilled readers use comparable cognitive processes and display similar eye movement patterns during reading. This thesis investigates individual differences in skilled readers’ eye movements across three experiments. We consider multiple cognitive processes that are active during reading, ranging from fine-grained letter and word level processing to higher-level comprehension tasks. Consistent findings suggest that the most stable predictors of individual differences in skilled readers’ word reading processes are skills related to lexical proficiency (such as vocabulary). Readers with greater lexical proficiency generally read more quickly than readers with low quality lexical representations. This difference is facilitated by faster word identification processes, especially when encountering less frequent words. In terms of encoding the position and identity of specific letters within words, consistent with findings from previous literature, skilled readers generally have a strict mechanism for encoding letter identities but are more flexible when encoding letter positions. Our results indicate that the flexibility of letter position encoding varies very little within groups of skilled readers, and as such appears to reach (near) maturation once reading skills are fully developed. A final experiment determined that skilled readers are able to adapt their reading strategies to different comprehension demands, and generally aim to read more quickly over time. Less skilled comprehenders appear to reach a threshold for how much they can speed up, even when comprehension demands are low. An important finding that arose within this thesis was that two often used offline comprehension tests, the NDRT and WIAT-II comprehension tests, never loaded together in principal components analyses, and as such appeared to measure distinct underlying constructs. As a result, we reveal a Jingle fallacy, an issue that occurs when two instruments are falsely assumed to measure identical constructs because they share a name. Findings and implications for future research practice are discussed.