Conjoint Analysis is a popular method in marketing research that is mainly used for product development. In a quasi-experimental setting product features (attributes) like brand, price or packaging are varied and the resulting synthetical products are evaluated by a sample of subjects. The major aim is to estimate (i) the attractivity of the products, (ii) the impact of the features for the product attractivity, and (iii) the market share. Estimating the attractivity is equivalent to scaling the products, and therefore we refer to (i) as the Scaling Problem and to (ii) as the Impact Problem. In most cases ANOVA-type of analyses are performed on both an individual and an aggregate level. The major disadvantages of this approach are, that the dependencies between the responses are not considered, and further that homogeneity of the subjects is assumed. For binary data from pairwise comparisons we propose to make use of a class of models from Item-Response-Theory (IRT), namely the Rasch Models. The Rasch Model (RM) was developed for the analysis of data from mental tests. It is a model for multivariate data that allows to model subject heterogeneity by the specification of corresponding parameters. The Linear Logistic Test Model (LLTM) allows to restrict the parameters of the RM, and here it is used for estimating the Scaling and the Impact parameters. The approach is applied to data from research on the objective determinants of subjective perception of car engine noise. The data analyses show that conventional Conjoint Analysis and the IRT-approach differ substantially in the estimates for product attractivity and market share. Modelling with and without subject variability makes a difference. The estimates from the models containing subject parameters are assumedly more reliable and consequently make a better basis for marketing decisions.