Abstract: On the basis of discussing general problems in cartographic generalization, this pa-per analyses the advantages and disadvantages of two technical lines for automatic general-ization of contour lines. The author suggests that it is possible to get faster and better general-ization results if we simulate the faster intelligence of human experts in program designing,retrieve geomorphological structural in formation using the input data of 2-dimensional con-tour lines,and directly derive and output the generalized 2-dimensional contour lines. The problems of artificial intelligence include:ordering of randomly digitized contour line data,recognition of important feature points on the contour lines,determination of positive or neg-tive geomorphological features, selection of valley points, the automatic tracing of valley lines, the calculation of importance indices of valleys, automatic measurement of distances between valleys, recognition of symmetric saddles,generation of generalized contour graph-ics, etc. In addition to the basic principles,data structures,and concrete algorithms,this paper finally shows the samples of generalization of contour lines from 1:10000 scale to 1:50000.