Data collected in the biomedical and social sciences by means of questionnaires is in most instances qualitative in nature. Such data, typically set out in the form of (multi-dimensional) contingency tables, is usually subjected to hypothesis testing in order to assess the interrelationships between the questions. Prior to undertaking confirmatory procedures, we argue that exploratory techniques should be used to gain a "feel" for the data. Correspondence Analysis (an exploratory data analysis procedure) and Log-linear Model building (a confirmatory data analysis procedure) are discussed before an investigation is undertaken to ascertain whether they can be used in conjunction. We found that correspondence analysis : (i) detects questions that are "strictly" independent/unrelated, (ii) detects pairwise relationships between questions (2-factor interactions) and thus can be used to suggest a splitting of large data sets into two or more subsets of questions that are independent, each of which can be analysed separately, and (iii) cannot be used to select log-linear models in general because it does not detect higher order interactions.