The improved Carbohydrate Structure Generalization Scheme has been developed for the simulation of 13C and 1H NMR spectra of oligo- and polysaccharides and their derivatives, including those containing non-carbohydrate constituents found in natural glycans. Besides adding the 1H NMR calculations, we improved the accuracy and performance of prediction and optimized the mathematical model of the precision estimation. This new approach outperformed other methods of chemical shift simulation, including database-driven, neural net based and purely empirical methods and quantum-mechanical calculations at high theory levels. It can process structures with rarely occurring and non-carbohydrate constituents unsupported by the other methods. The algorithm is transparent to users and allows tracking used reference NMR data to original publications. It was implemented in the Glycan-Optimized Dual Empirical Spectrum Simulation (GODESS) web service, which is freely available on the platform of the Carbohydrate Structure Database (CSDB) project (http://csdb.glycoscience.ru).