Abstract: The conceptual data model of a GIS directly represents people's understanding of the external world.Currently,all kinds of GIS data model can be generalized to two types:Topological Relation Data Model and Entity-oriented Data Model.These two kinds of conceptual data model have their advantages and disadvantages respectively.Topological relation data model has such advantages that structure is compact,topological relation is explicit,and it is efficient in network analysis and topological query,but has also some disadvantages:① being inefficient to operate one complete geographical entity;② being unable to efficiently represent complex geographical entities;③ being difficult to query those complete geographical entities;④ being difficult to update local geographical region and maintain GIS software.Compared to Topological Relation Data Model,topological relation is not a basis of data organization any more,and topological geometric elements are not also used as basic units of data storage and operation.Entity-oriented data model organizes data based on complete and independent geographical objects with realistic sense.It has many obvious advantages:that it is easy to manage,query and modify entities,easy to make spatial analyses and efficient to represent those complex geographical entities,and its data organization mode fits the manners people observe the exterior world and users can naturally understand and accept it.At the same time,based on this model,GIS software is easily maintained and expanded.Nevertheless,it has some disadvantages:①Topological relation needs to be built according to actual demands; ② It is inefficient in dynamically section and network analysis; ③Those common points and arcs will be repeatedly stored; ④Operations can only be done on those complete geographical entities and are unable to be efficiently executed on the level of geometric elements.These two kinds of data model have some common disadvantages:①Their layers cannot satisfy the holistic feature requirements of the objective world; ②They do not pay enough attention to semantic relation between two objects; ③They can not efficiently represent fuzzy and uncertain phenomena. To represent the objective world better,this paper gives a new understanding of geographical space:①Geographical space is one relative object-based space and one geographical object group; ②Geographical space is one 3D space; ③Geographical space is one temporal and multi-scale space.Based on the new understanding,the authors present five important points of designing holistic GIS data model.① Geographical space must be represented as one geographical object group according to object-oriented thinking.② Semantic relation between objects should be paid more attention,and geometric,attribute and time information in GIS should be handled with same importance.③ Holistic GIS data model has to give more functions to layer (especially to compound layer) such as adding and removing geographical object,query and spatial analysis,etc.④Object model and field model can be unified with object-oriented method.⑤A part of geographical objects,must have 3D information such as elevation value.As a result,one design of holistic GIS data model is presented at the end of this paper.