We study how people evaluate and aggregate the attributes of naturalistic choice objects, such as movies and food items. Our approach applies theories of object representation in semantic memory research to large-scale crowd-sourced data, to recover multiattribute representations for common choice objects. We then use standard choice experiments to test the predictive power of various decision rules for weighting and aggregating these multiattribute representations. Our experiments yield three novel conclusions: 1. Existing multiattribute decision rules, applied to object representations trained on crowd-sourced data, predict participant choice behavior with a high degree of accuracy; 2. Contrary to prior work on multiattribute choice, weighted additive decision rules outperform heuristic rules in out-of- sample predictions; and 3. The best performing decision rules utilize rich object representations with a large number of underlying attributes. Our results have important implications for the study of multiattribute choice.