The ULTRA-WIDE BAND (UWB) signals transmit a large amount of information over a short distance with low power and the signals reflected by the inspected materials can be obtained without contacts of the materials. As a result, the reflected UWB signals offer us one potential contactless material identification or classification tool. In this paper, we study the UWB signals collected in a series of liquid material classification tests. We apply the spectral clustering algorithms to group the UWB signals into some desired number of classes. The results demonstrate the potential of UWB based material classification. The data preprocessing and clustering algorithm selection problems are explored as well.