Whereas previously the output of HIV resistance tests has been based on therapeutically arbitrary criteria, there is now an ongoing move towards correlating test interpretation with virological outcomes on treatment. This approach is undeniably superior, in principle, for tests intended to guide drug choices. However the predictive accuracy of a given stratagem that links genotype or phenotype to drug response is strongly influenced by the study design, data capture and analytical methodology used to derive it. For genotyping, the most widely used resistance tool in clinical practice, these considerations are further complicated by the range of mutational patterns present in the treated population. There is no definitively superior methodology for generating a genotype-response association for use in interpreting a resistance test, and the various approaches used to date all have their strengths and weaknesses. This review discusses the processes involved in constructing such tools, with particular emphasis on establishing validated mutation score rules, and examines the key issues and confounding factors that influence predictive accuracy outside the originating dataset. Since the size of the sample is a key influence on the statistical power to determine an effect, it is hoped that a greater understanding of the influence of study design and methodology will assist the development of standardized outcome measures and reporting formats that allow data pooling at the international level.