A novel method is proposed for deciphering experimental tandem mass spectra. A large database of previously resolved peptide spectra was used to determine "neighborhood patterns" for each peak category: C- or N-terminus ions, their dehydrated fragments, etc. The established patterns are applied to assign probabilities for new spectra peaks to fit into these categories. A few peaks often could be identified with a fair confidence creating strong "anchor points" for De Novo algorithm assembling sequence subgraphs. Our approach is utilizing all informational content of a given MS experimental data set, including peak intensities, weak and noisy peaks, and unusual fragments. We also discuss ways to provide learning features in our method: adjustments for a specific MS device and user initiated changes in the list of considered peak identities.