This thesis develops a quantitative method for forensic authorship attribution. The principal constraint, that the method be scientific according to the Daubert criteria, necessitates that the conclusions drawn about authorship problems must be made to a known degree of certainty. In response, the theoretical part of the thesis establishes the criteria for a sound method in authorship attribution as relying on valid, reliable markers of authorship and the development of an explicit and specific sampling strategy. The main empirical part of the thesis draws potential markers of authorship from the literature and tests them against a specially constructed General Authorship Corpus. The resulting battery of reliable markers of authorship includes word and sentence length statistics and word-frequency measures. A series of worked examples with decreasing number of texts demonstrates the method and tests its limits, showing positive attributions where possible and no false attributions even when comparison data is limited. In addition to the development and application of the battery of valid, reliable markers of authorship, the role of stylistic idiosyncrasies in attribution is discussed and developed as a secondary strategy. Possibilities for the statistical presentation of results are considered and a Bayesian approach is proffered as the most desirable