This paper presents a formal approach to analyzing behavioral sequences produced by computer users. Following this approach, the behavioral sequences are represented as square transition matrices, distances between the individual matrices are calculated, and the resulting distance matrix is scaled multidimensionally. The emerging scales reflect differences in the users' problem solving behavior and can be used as a guideline for detecting behavioral aspects influencing quality of task solution. The method was exemplarily applied to two different kinds of behavioral data, the first reported by Upmeyer, Meseke & Enzmann (1988) and the second by Eyferth, Widowski & Schmid (1991).