Decisions such as selection of the 'best' person for the job or the most effective point at which to conduct performance training are based to a large extent on personal judgments formed from an analysis of 'soft' data. In the behavioural sciences, soft data can often take the form of self-report responses to a questionnaire, which is very susceptible to 'noise' such as missing or incorrect item responses. Neural nets have been shown to be an effective method of assessing 'noisy' data. This paper describes a systematic comparison of a neural net analysis with the more traditional statistical factor analysis. It was shown that a correctly developed neural net provided a tool that was more sensitive to the complex interrelationships between items on a measurement instrument (questionnaire) than was obtained with traditional statistical techniques. Development of a new heuristic similar to network 'pruning' allowed identification of 'good' and 'bad' questions based on input-to-output sensitivity rather than subjective judgment. Results of the study indicate that a neural net approach to construct validation could provide a more sensitive measure of specified characteristics and abilities when used as an aid in making vital decisions regarding personnel selection and performance training.