A database of 3D structures of natural metabolites has been developed called 3DMET. During the process of structure conversion from 2D to 3D, we found many structures were misconverted at chiral atoms and bonds. Several popular converters were tested in regard to their conversion accuracy. For verification, three canonical strings were also tested. No procedure could satisfactorily cover all the structures of the natural products. The misconverted structures had to be corrected manually. However, a nonnegligible number of mistakes were also observed even after manual curation, so a self-checking system was developed and introduced to our work flow. Thus, the 3D structures in our 3DMET database were evaluated in two steps: automatically and manually. The current version includes most of the natural products of the KEGG COMPOUND collection [ http://www.genome.jp/kegg/compound/ ] and is searchable by string, value range, and substructure. 3DMET can be accessed via http://www.3dmet.dna.affrc.go.jp/ , which also has detailed manuals.