Generalizability theory (GT) postulates that there are infinite sources of error variation in any measurement. Generalizability analysis is used to reduce and control this measurement error. To date, there has been no study employing GT in analysing taekwondo, despite the increase of observational studies in combat sports. Through applying GT to six random combat bouts within the Olympic Games in London 2012, the aims of this study are: (a) to determine whether the bouts selected are representative enough for their behaviour to be generalizable; (b) to determine the intra and interobserver agreement and reliability; and (c) to assess the accuracy of the exhaustive and mutual exclusion of each category included in each criterion of the categorical system. Method: Six taekwondo bouts were randomly selected in order to analyse the athletes' technical/tactical behaviour. Three observers were responsible for viewing and analysing the videos using a consensual concordance strategy. Results and discussion: High values for intra and interobserver agreement were found in Kendall's tau and Cohen's kappa indexes, in addition to the GT. The results showed that the categorical system was composed of exhaustive and mutually exclusive (E/ME) categories and that the sample was representative enough to generalize the behaviour shown.