This thesis proposes a formal model of interaction in games, to be used as tool for game analysis and game testing. The model allows a quantification of interaction by looking at the low-level structure and patterns in game-controller input. The game-controller input is modelled using discrete-time, discrete-space Markov chains, and information theory is used to quantify the mismatch between the models prediction and the actual user input. The model uses game-agnostic game controller data as its input, which is the lowest common denominator for a large class of games (almost all game console games, most PC games). The models are trained dynamically on-the-fly for each individual play session. This allows performing individual analyses of players interactions, while still retaining an approach that is very general and can be used with different games without modification. To adapt to new play situations quickly, the used models are only based on data from the last couple of seconds or minutes. This can lead to the problem that not enough samples may be available to confidently estimate all dynamic model parameters. This problem is mitigated by considering the full probability distribution of each parameter instead, using a beta distribution. This work contributes to the understanding of interaction in games, modelling of raw user input and quantifying the model output using information theory. The described approach has been implemented in software and preliminary results from a prestudy are available. In this exploratory prestudy, the post hoc analysis of nine different games from various genres revealed a number of interaction patterns. One of the observed patterns is routinization, a process in which an action is performed repeatedly until it is executed almost unconsciously. Research in this field, based on this thesis, has been performed in cooperation with Martin Pichlmair from the IT University Copenhagen, and a workin- progress paper is to be published in the proceedings of the ACM SIGCHI Annual Symposium on Computer-Human Interaction in Play (CHI PLAY) [Wallner et al., 2015].