Abstract: The characteristics of existing approaches to the selection of streets are summarized based on an analysis of the literatures.The metric information,topological information and thematic information are calculated for street networks.From an artificial intelligence perspective,the generalization process can be seen as an optimization problem.A progressive approach to the selection of streets base on information loss is proposed.This approach minimizes the information loss.An experiment was conducted on a partial map of an urban settlement from the compilation specifications for 1∶25 000,1∶50 000 topographic maps.The Results show that our proposed approach can select the streets progressively and dynamically.