A typical distribution network contains hundreds of medium-voltage (MV) feeders and ten thousands of low-voltage (LV) feeders. This work introduces an efficient taxonomy approach that combines cluster analysis with discriminant analysis to identify statistically representative MV and LV feeders in the west Australian context. Quadratic discriminant functions have been extracted and can be used as a feeder classifier for any feeder in this distribution system. The representative feeder sets provide rigorously validated test cases for the evaluation of smart grid technologies.