The distribution network impact assessment of changes in load behaviors or new technology deployments requires a rigorous understanding of the features of the network itself. In a typical distribution network, there will be hundreds of high voltage (HV) feeders and ten thousands of low voltage (LV) feeders. This paper presents a method to combine cluster and discriminant analysis techniques to identify a small number of statistically representative or prototypical feeders that capture the key features of a distribution network. The proposed method is readily transferable to other systems. As an illustration the paper presents a representative HV feeder set for an existing network and a feeder classifier based upon quadratic discriminant functions. Representative feeder data will be typically used as an input to Monte Carlo models to assess the impact of technology changes driven by Smart Grid deployments or load changes due to distributed renewables, air conditioning or electric vehicle uptake.