This study presents a power network reduction algorithm based on coherency. Since electrically close generators tend to swing together upon a disturbance, we determined coherent generator groups using their mutual electrical proximity. Rank correlation coefficient was defined as a measure for the degree of this proximity. These coefficients were calculated from the bus admitance matrix of the system. This measure could be used to determine the degree of the proximity of any two buses in the system also, hence served as a network partitioning algorithm. The order of partitioning could be chosen so as to one of the partitions include a pre-defined study area. We used Kron reduction to eliminate load buses. The algorithm was tested on a 39-bus system and Turkish 380 kV system. Power flows through the lines and generator swing curves in the reduced network model were compared with those in the full network model. Results are satisfactory. We found that an electrical proximity based coherency technique was proper for power network reduction.