A methodology for designing a Bonner sphere spectrometer was proposed based on the singular value decomposition (SVD) and information entropy theory. First, features of Bonner spheres were classified based on SVD, and then, the rationality of the selected spheres was analyzed. Second, the information entropy is used as the quantitative evaluation standard; combined with the unfolding method, the contribution of each Bonner sphere on the information entropy of the unfolded spectra was quantitatively analyzed. The spheres that have high influence on the information entropy of solution spectra can be identified, and then, the information entropy variation of the solution spectra can be suppressed by reducing the count fluctuation of these spheres. According to the proposed method, an optimized spectrometer was developed and applied to measuring 252Cf spectra.