Abstract: Aiming at the issue of using multi-purpose geochemistry data to study the classification of quaternary sediment, recognition model based on probabilistic neural networks was put forward and the method and steps of selection of geochemical index, index normalized and the setting and training of neural networks were given. Eight kinds of genetic quaternary sediments were distinguished availably by probabilistic neural networks in Songnen Plain，the central and western Jilin Province, and the problems for the genesis of quaternary sediments was solved well in this area, which showed that the method was of strong nonlinear recognition ability. The recognition ability of the probabilistic neural networks model for Quaternary sediments is higher than that of traditional multivariate statistical analysis. And this model with simple structure provided can be trained fast, thus alsois the valid method using multi-purpose geochemistry data to recognize the quaternary sediment types.

Key words: Quaternary sediment, recognition, multi-purpose geochemical data, probabilistic neural networks, the central and western Jilin Province