Symbolic data represents a general form of classical data. There has been a highly focused research on the analysis of symbolic data in recent years. Since most of the future applications involve such general form of data, there is a need to explore novel methods to analyze such data. In this paper we present two simple novel approaches for the classification of symbolic data.1 In the first step, we show the representation of symbolic data in binary form and then use a simple hamming distance measure to obtain the clusters from binarised symbolic data. This gives the Class label and the number of samples in each cluster. In the second part we pick a specific percentage of significant data samples in each cluster and use them to train the Adaptive Auto-configuring neural network. The training automatically builds an optimal architecture for the shown samples. Complete data has been used to test the generalization property of the RBF network. We demonstrate the proposed approach on the soybean bench mark data set and results are discussed. It is found that the proposed neural network works well for symbolic data opening further investigations for data mining applications.