Ability to measure value from the customer's point of view is central to the determination of market offerings: Customers will only buy the equivalent of perceived value, and companies can only offer benefits that cost less to provide than customers are willing to pay. Conjoint analysis is the most popular individual-level value measurement method to determine relative impact of product or service attributes on preferences and other dependent variables. This research focuses on how value measurement can be made more accurate and more reliable by measuring the relative influence of selected methodological variations on performance in prediction and on stability of value structure, and by grouping customers with similar value structure into segments which respond to product stimuli in a similar manner. Influences of the type of attributes included in the conjoint task, of the factorial design used to construct the product profiles, of the type and form of model, of the time of measurement, and of the type of cluster-based segmentation method, are evaluated. Data was gathered with a questionnaire that controlled for methodological variations, and with a notebook computer as the measurement object. One repeated measurement was taken. The study was conducted in two phases. In Phase I, influences of methodological variations on accuracy in prediction and on respective value structure were examined. In Phase II, different cluster-based segmentation methods--hierarchical clustering (HIC), non-hierarchical clustering (NHC), and fuzzy c-means clustering (FUC)--and according conjoint models were evaluated for their performance in prediction and in comparison with individual-level conjoint models. Results show the best models for a variety of design parameters are traditional individual-level, main-effects-only conjoint models. Neither modeling of interactions, nor segment-level conjoint models were able to improve on prediction. Best segment-level conjoint models were obtained with a fuzzy clustering method, worst models were obtained with k-means and the most fuzzy clustering approach. In conclusion, conjoint analysis reveals itself as a reliable method to measure individual customer value. It seems more rewarding for improvement of accuracy in prediction to apply repeated measures, or gather additional data about the respondent, than to attempt improvement on methodological variations with a single measurement.