When most designers set out to develop a new product, they solicit feedback from potential consumers. These data are incorporated into the design process in an effort to more effectively meet customer requirements. Often these data are used to construct a model of consumer preference capable of evaluating candidate designs. Although the mechanics of these models have been extensively studied, there are still some open questions, particularly with respect to models of aesthetic preference. When constructing preference models, simplistic product representations are often favored over high fidelity product models in order to save time and expense. This work investigates how choice of product representation can affect model performance in visual conjoint analysis. Preference models for a single product, a table knife, are derived using three different representation schemes: simple sketches, solid models, and three dimensional (3D)-printed models. Each of these representations is used in a separate conjoint analysis survey. The results from this study show that the choice model based on 3D-printed photopolymer prototypes underperformed. Additionally, consumer responses were inconsistent and potentially contradictory between different representations. Consequently, when using conjoint analysis for product innovation, obtaining a true understanding of consumer preference requires selecting representations based on how accurately they convey the product details in question.