A new application of K-NN was implemented to classify various substances using measured UWB channel complex coefficients. This study provides a low-cost system to detect different substances by applying a data mining technique to the collected frequency traces. Measurements of UWB channel coefficients were conducted for two types of crude oil and ocean water, independently contained within 5 and 10-l PVC containers. These containers were located as the device under test between directional transmitting and receiving ultra wideband antennas and were both connected via the Vector Network Analyser to measure their channel frequency responses over frequency bands of 300 MHz to 8 GHz. Magnitude differences were subsequently taken between the various baselines (empty PVC containers) and filled containers of the crude oil types (denoted A and B) or ocean water. Random noise was also subsequently added to this data to test the robustness of the classification method. After applying K-NN to the collected data it was found that our classification results were 100% for the sets of 5 l and 10-l crude oil samples but reduced to 56.66% and 60.66% when the random noise with standard deviations of one was added to the recorded data.