Although there is enormous evidence that reference levels influence preferences, conjoint models, one of the most successful marketing research tools, assume that preferences depend on the absolute levels of attributes. In this paper we investigate the relevance of reference effects in two settings, compositional or self-explicated models in experimental studies 1 and 2, and decompositional or choice-based models in experimental study 3. In particular, we introduce a simple modification of the traditional self-explicated conjoint model which permits dependence of preference on reference levels. By eliciting gains and losses from expectations the model is adaptable to changes in respondents' reference points, which the traditional model is incapable of. Reference options are found to clearly affect subject choices in studies 1 and 2. In addition, the reference dependent self-explicated model is found to offer useful predictions when reference points are manipulated in study 1, and improve on predictions of its traditional counterpart when reference points are measured in study 2. In contrast, in study 3, the choice-based model’s diagnostics and predictions are found to be robust to reference point manipulations. Taken together, these results suggest that the self-explicated model is more suited than the choice-based model to understanding and predicting how respondents make judgments relative to reference points because reference points and gains and losses from reference levels are more salient in the self-explicated model. We discuss implications for managers constructing conjoint models in product-market settings wherein reference points are changing due to new product introductions or marketing efforts.