Abstract: The mass data in the large scale database has mixed attributes, namely, the mixed data of symbolic data and numerical data, and the quantity of data is complex and difficult to distinguish. Traditional algorithms often ignore the correlation between the two attributes, the calculation is complex, the clustering speed is slow, the effect is poor. A numerical study of mixed database clustering in large-scale data clustering algorithm based on the traditional algorithm, firstly in order to reduce the high complexity, from reasonably extracting large-scale databases of multiple small data sets, all natural clusters contain database on small data sets; similarity matrix similarity is defined to construct small data set based on the then, the similarity data symbols and numerical data calculation; integration, the final result of the sample clustering updated mean and classification, clustering of numeric data mixed in large databases. Simulation results show that the proposed algorithm can get a better clustering result, and is suitable for large scale data clustering processing. The algorithm has a fast calculation speed, a relatively small amount of computation, and a low error rate.