The electrocardiogram (ECG) is a graphical representation of the electrical activity of the heart. The 12-lead ECG shows this activity in 12 "views" called "leads", relative to the location of sensors attached to the body surface. The ECG is a routinely applied cost effective diagnostic medical test, utilised in healthcare settings around the world. Although more than three hundred million ECGs are recorded each year, correctly interpreting them is considered a complex task. Failure to make correct interpretations can lead to injury or death and costs vast sums in litigation payments. Many automated attempts at interpreting ECGs have been implemented and continue to be developed and improved. Despite this, automated methods are still considered to be less reliable than expert human interpretation. As ECG interpretation is both a cognitive and visual task, eye-tracking holds great potential as an investigative methodology. This thesis aims to identify any cues in visual behaviour that may indicate differences in subsequent ECG interpretation accuracy. This is the first work that uses eye-tracking to analyse how practitioners interpret ECGs as a function of accuracy. In order to investigate these phenomenon, several experiments were carried out using eye-movements captured from clinical practitioners that interpret ECGs as part of their usual clinical role. The findings presented in this thesis have advanced research in the understanding of ECG interpretation. Specifically: Clinical history makes a difference to how people look at ECGs; different gaze patterns are often found in accurate and inaccurate interpretation groups. Grouping data to account for within ECG lead behaviour (eye-movement patterns within a lead) is more revealing than analysis at the level of the lead (eye-movements between leads). Findings suggest analysing visual behaviour at this level is crucial in order to detect behaviour in ECG interpretation. Further to this the thesis presents eye-tracking techniques that can be applied to wider areas of task performance. These methods work over complex stimuli, are able to deal post hoc with differently sized groups and generate appropriate areas of interest on a stimulus. These methods detect important differences in eye-movement behaviour between groups that are missed when applying standard inferential statistical techniques.