The purpose of this paper is to provide a design method for exhibition spaces that incorporates knowledge of spatial preferences. This knowledge is based on a previous paper where probabilistic models were used to represent the correlation between preference and spatial elements used in exhibition spaces. This paper serves to demonstrate these models as design knowledge. We have executed the probabilistic reasoning of Bayesian networks on the previous models, and deduced division- patterns for exhibition spaces and combinations of spatial elements with an expected high probability of preference. New schemes for exhibition spaces were then developed, based on these reductions.