Capturing meaningful interactions between a user and an application is useful. A meaningful interaction is an interaction that causes a change in state of one of the participants. Meaningful interaction capture is well established for console based applications. No techniques exist that capture only the meaningful interactions with a graphical interface. Data collected by such a system could be used for a variety of applications, such as HCI studies, authorship identification, cognitive modelling and screen recording. A methodology for capturing the meaningful interactions between a user and a graphical application is described. An implementation of this methodology has been developed, together with a supporting tool-set. A new corpus consisting of captured interactions between users and two applications with contrasting graphical interfaces has been collected and published. This corpus is analysed and used for authorship attribution. The use of this interaction capture system is evaluated as a high compression screen recorder. By using the interaction capture system as a screen recorder it is shown that the size of captured files are an order of magnitude smaller than the equivalent file created by a video based screen recorder. Analysis of the captured corpus gives an overall accuracy of 83 percent when predicting the author of a stream. This is significant, showing that the way people interact with an application is unique.