The Self-Organizing Map (SOM) is widely accepted as a data visualization and cluster model for its ability to map high dimensional data in a low dimensional output space according to the data?s similar features. However, this mapping process is time consuming and a large amount of iterations are needed in order to increase the accuracy of the data representation. This paper describes how to apply the RGB color model to the initialization of the SOM neurons. The major feature is that the distribution of the neurons is closely related to the data distribution during the initialization of SOM. Therefore the iterations are greatly reduced and efficiency and accuracy of SOM are much improved. To evaluate our approach against traditional approaches we have conducted an experiment. The initial results show that the color model based 3-D SOM is very promising in the practical application.