Magic-angle-spinning solid-state (13)C NMR spectroscopy is useful for structural analysis of non-crystalline proteins. However, the signal assignments and structural analysis are often hampered by the signal overlaps primarily due to minor structural heterogeneities, especially for uniformly-(13)C,(15)N labeled samples. To overcome this problem, we present a method for assigning (13)C chemical shifts and secondary structures from unresolved two-dimensional (13)C-(13)C MAS NMR spectra by spectral fitting, named reconstruction of spectra using protein local structures (RESPLS). The spectral fitting was conducted using databases of protein fragmented structures related to (13)C(α), (13)C(β), and (13)C' chemical shifts and cross-peak intensities. The experimental (13)C-(13)C inter- and intra-residue correlation spectra of uniformly isotope-labeled ubiquitin in the lyophilized state had a few broad peaks. The fitting analysis for these spectra provided sequence-specific C(α), C(β), and C' chemical shifts with an accuracy of about 1.5 ppm, which enabled the assignment of the secondary structures with an accuracy of 79 %. The structural heterogeneity of the lyophilized ubiquitin is revealed from the results. Test of RESPLS analysis for simulated spectra of five different types of proteins indicated that the method allowed the secondary structure determination with accuracy of about 80 % for the 50-200 residue proteins. These results demonstrate that the RESPLS approach expands the applicability of the NMR to non-crystalline proteins exhibiting unresolved (13)C NMR spectra, such as lyophilized proteins, amyloids, membrane proteins and proteins in living cells.