Geographical processes are scale dependent and numerous applications in GIS require changing degrees of detail or scale

thus geographic information should be made available at multiple levels of abstraction. In this sense

people are now beginning to realize the lack of generalization functionality with respect to the development of value-added products from the initial database and the update of existing databases

particularly when multiple scales are involved. Geographical information generalization is a new challenge to technical progress of GIS. This paper focuses its basic meaning and realization firstly

then introduces a new automatic algorithm based on fractal analysis

which is proposed by the authors

and discusses the further problems of generalization in GIS lastly.