Applied eye-tracking has been extensively used for the study of psychological processes. More recently, some researchers have used this technique to study the interaction between people by tracking and analyzing eye-movements of two persons synchronously. However, this is generally accomplished by observing people in simple controlled settings. In this thesis, we use a similar methodology, dual eye-tracking, to study people in more natural, semantically richer, tasks with the aim of identifying dual eye-movement patterns which reflect collaborative processes. Eye-tracking, and more globally eye-movements analyses, is a complex domain involving several methodological issues, which have not yet been satisfactorily solved. This work is also an attempt to offer solutions to several of these issues. The first part of this thesis is dedicated to the improvement of the methodology of eye-tracking data analysis. We present several developments pertaining to the general methodology of eye-tracking. More specifically, we identify problems and offer solutions to the following aspects: fixation identification, systematic position errors correction and hit detection. We also tackle more specific questions concerning the method dual eye-tracking. We present issues that arise in dual eye-tracking data collection and analysis and propose some solutions. On the technical side, we deal with the question of the synchronous recording of two streams of eye-movements and on the analytical side, we extend gaze cross-recurrence, a measure of eye-movements coupling, to complex realistic collaborative tasks. The second part is devoted to experimental studies of collaboration through the use of dual eye-tracking methods. We first present four exploratory studies which allowed us to set up the stage by identifying interesting phenomena and experimental difficulties. The main results of these experiments revolve around the relationship between gaze and speech. In these respects, we extended some results found in the literature to more natural settings and we developed a computational model to make actual predictions about dialogue and visual references. Finally, the main study of this thesis is about computer program understanding. This study is composed of two experiments: a solo programming experiment, from which we identified gaze patterns of a single programmer and a pair-programming task in which we explored how gaze patterns during program comprehension are affected by collaboration.