Space for the exhibition is often designed through the composition of different geometrical elements with a strong figure-ground relationship. It is composed of rooms and space in between the rooms. Composition equilibrium, ambiguous interesting spatial quality, and functionality of each exhibition area become three crucial criteria of the space. However, aesthetic of composition and spatial quality are both hard to define, which I try to explore in the generative design approach. In the project, I try to design the exhibition space of surprises through the composition of several geometrical elements innovated by Kandinsky. Is it possible to use certain metrics as bait for the computer to find “surprising” designs for us? The definition of surprising space quality is defined by the standard deviation of the visual area on the paths from the entrances to the exits and the difference between two neighboring positions. Through the generative process, we might understand some subjective conceptions better and the computer could be capable to help us find subjectively high-performance designs.