The abundant observation of chemical fragment information for molecular complexities is a major advantage of biological NMR analysis. Thus, the development of a novel technique for NMR signal assignment and metabolite identification may offer new possibilities for exploring molecular complexities. We propose a new signal assignment approach for metabolite mixtures by assembling H-H, H-C, C-C, and Q-C fragmental information obtained by multidimensional NMR, followed by the application of graph and network theory. High-speed experiments and complete automatic signal assignments were achieved for 12 combined mixtures of (13)C-labeled standards. Application to a (13)C-labeled seaweed extract showed 66 H-C, 60 H-H, 326 C-C, and 28 Q-C correlations, which were successfully assembled to 18 metabolites by the automatic assignment. The validity of automatic assignment was supported by quantum chemical calculations. This new approach can predict entire metabolite structures from peak networks of biological extracts.