This thesis presents a multi-method approach to the investigation of lexical opposition, taking into account central as well as peripheral opposites. One of the distinguishing factors of the present research is the investigation of a wide variety of opposite pairs (e.g. converses) alongside the canonical gradable adjectival antonym pairs which have thus far been the focus of most of the research on lexical opposition. Three main questions concerning lexical opposition are addressed from a data-driven perspective. Firstly, the question of which factors determine the antonymic strength of a pair of opposites. Secondly, whether there is a clear distinction between good and bad opposites or whether the internal structure of the category of antonymy can be described in term of prototype theory. Lastly, whether antonymy is a relation between cognitive concepts (Murphy & Andrew 1993) or between the words which encode these concepts (Gross et al. 1989). This study employs a wide range of methodologies: corpus analysis, judgement tasks, attribute listing and behavioural lexical decision tasks in English and German. This allows for the detection of discrepancies between the results of the separate investigations which provide the starting points for the analyses. The data shows that the existing criteria for ‘good’ opposites influence a pair’s antonymic strength to different degrees. All factors which have previously been put forward with the addition of those which are proposed in this thesis are integrated into a holistic approach based on the theory of cognitive concepts and cognitive models (Lakoff 1987). The evidence shows strong support for antonymy as a prototype category with antonym pairs distributed along a continuum of antonym canonicity on which each pair’s place is determined by both its antonymic strength (which is influenced by a certain distribution of factors) and its associative strength. The contrastive data provides the strongest evidence for antonymy as a conceptual relation since opposite pairs in both languages exhibit very similar behaviour.