Abstract: The dynamic mode decomposition (DMD) method improved by embedded dimension and the proper orthogonal decomposition (POD) method are used to analyze the random wind pressure field of a large-span flat roof. The results show that the fuzzy dynamic features hidden in the wind tunnel test data can be discovered by using the Takens embedding theorem to increase the spatial dimension of the original snapshot matrix. The modal distributions of POD and DMD can reflect the main characteristics of typical vortices in a fluctuating pressure field. However, each mode of POD contains the information of multiple frequencies, which to a certain extent makes the mode of POD become the fluctuating coupling of multiple frequency bands. The mode decomposed by DMD method is a single-frequency mode and the stability of each mode can be obtained. At the same proportion, POD modes contain more energy than DMD modes. But when the same proportion of modes are used to reconstruct the fluctuating wind pressure field, the reconstruction results of DMD method can better describe and fit the local characteristics of the original fluctuating pressure field than those of POD method. This is because DMD method reconstructs the pressure field directly, while POD method mainly reconstructs the energy field. Therefore, the DMD method has more advantages in revealing the spatial-temporal evolution characteristics of a random wind pressure field.