Network pattern mentioned in this paper is referred to the geographical layout and structure of a network, which is related to the connection, direction, and combination features of roads in a road network. A quantitative method is proposed in this paper to classify patterns of networks, through which a network could be identified quantitatively to be one of the three standard patterns, i.e., grid network, circle+radial network and tree-patterned network. Metric distances of shortest paths are taken as the main features of the networks and are described by parameters through Generalized Extreme Value (GEV) fitting. The criteria for pattern classification were established according to the cluster analysis of the parameters calculated from a set of trial networks. Six real networks were calculated using the method and their patterns are identified according to the proposed criteria. It turned out that the method could capture the features of the network patterns well. This method may set a threshold of the more general and deep studies of network pattern classification, which may offer help to road network planning and assessment.