Mineral information includes all kinds of relative metallogenic information. In order to extract comprehensive metallogenic prediction information, it's necessary to filter initial observation information to emphasize the factors which are most advantageous to metallogenic. Rough set can delete irrespective or unimportant attributes under the premises of no classification ability changing, without supplement information or prior knowledge. A new integrated predicion model based on Rough set theory is put forward in this research. The mineral information most advantageous to metallogenic from a great number of variables to achieve the optimization of variable structure and numerical interval is chosen. Based on the optimization combination, characteristic function is established for prediction. Combined with some conventional methods for deposit statistics, prediction, clustering means is applied to get the critical point for decision and quantitative charcterisic analysis is applied to predict the mineral resource by calculating the relation degree of are every geological cell. And eight geological cells are established as the cells advantageous to metallogenic. Results are basically in accord with practice, which shows availability of this method.