Two-dimensional arrays are the key components used in the coded aperture imaging, and directly influence the quality of reconstructed images. Unfortunately, up to now, there is not a sort of two-dimensional arrays with fine property of tomographic imaging and high photon collection rate. The authors use a new global optimization method, DIRECT algorithm, to design two-dimensional arrays, which is very useful when the object function is a “black box” and has a faster convergence rate than others. The goal is designing a sort of arrays with max sidelobe of their autocorrelation function no more than 1 and the filling rate as high as possible. Theoretical analysis and experimental results prove that the arrays possess better property of tomographic imaging and higher photon collection rate than the existing arrays.