Current BtoC recommendation services utilize consumers' purchased log as criteria for selecting information, yet it includes little information of the reason why he bought the items. Thus it is difficult to recommend the suitable information for each consumer.We have observed how each consumer judges his likes and dislikes on objects viewing them. We have modeled each consumer's evaluation process by relationships among physical features of objects, each consumer's subjective interpretations and preferences. Thus, based on the models, our system can estimate users' subjective evaluations and preferences from physical features of objects to perform a suitable recommendation.We have also build situated preference models according to the usage of the items. The recommendation system refers such situated preference models together with the subjective evaluation model as the consumer's decision-making model for selecting objects.