Abstract: The categorical elements of city graphics are street network, city blocks, buildings and other geo-features.But in the digital environment, generally building objects are stored in the database while the street objects or city blocks are not.Consequently such situation is unable to support the generalization processes.In fact, both streets and city blocks are complex objects.The former consists of at least left and right street side lines which are also composited of several street side segments.The latter can include several buildings and/or other geo-features.Therefore, it is necessary to introduce concept"street side line" for creating the street object itself as a complex object at one hand and such street side line can serve as incidencor between street and buildings located on the both sides of this street at the other hand.For generalization the following relations are to be established:(1)the main connective relations between principal streets;(2)the relation between street and city management boundaries;(3)the incidence relation between city blocks and street objects;(4)the inclosion relation between city blocks and geo-features.The generalization of city graphics should be realized by handling two major problems:(1)logical levels of generalization, (2) practically operational approaches for object selection and relation preserving.The logical levels of generalization should be divided into three hierarchies:street network generalization, city block generalization and buildings/other feature generalization.For street network generalization several methodological approaches can be chosen:for example, graph theory based approach, con-cept-geometrical approach, etc.In this paper the author has modified the Morgen's topological enco- ding method for network structure whose initial idea is to calculate the cardinalities as a sum of the adjacent node cardinality values for all non-peripheral nodes.The modification embodies in the following:the calculation of cardinalities carries out for all nodes including all peripheral ones and a further optimation in form of sum of weighted multiplications for all adjacent nodes: Cnew=∑ WiCi, where i denotes the number of adjacent nodes.This encoding idea is rather clear and the computing process is simple.Because the widening of major street symbols before city block generalization the entire city block graphics must be transformed by affine method.The next stage of city block generalization is merging the close located buildings.Beyond the dynamical decision tree and the other similar methods the author has provided a "gravitation principle" of building merging algorithm.Here the gravitation exists between the closed buildings in the direction of line connected their centroid points.The merging process consists of the following stages:(1)recognizing the close located buildings;(2)determining the centroid of the buildings related;(3)rotating the coordinate system in vertical direction connecting the two centroids; (4)connecting the related buildings in the following 6 versions:(a) strongest merging-establishing the minimal boundary rectangle;(b)stronger merging-creating the convex shell of the related buildings;(c)general merging-creating two half convex shells of outside parts relative to inter centroid part and connecting the upper and lower extreme points inside the centroid part respectively;(d) strong merging of the parts inter the centroids only, the outer parts retain unchanged;(e) medium merging of the parts inter the centroids only convexing the inside parts and connecting the closed located building sides;(f)weak merging the parts inter the centroids-recognizing the close located building sides and connecting them.