Abstract: Consistency, one of spatial data quality characteristics, plays an important role in such fields as spatial analysis, spatial query and spatial decision making.Only when spatial data is topological consistent, can GIS application obtain reliable quality.But GIS data is captured from various application sources in different time with different resolution, integration of multi-scale spatial data inevitably has inconsistent problems.Data handling, such as map generalization also destroys consistency.So it is necessary to find special strategy to preserve data consistency. In the process of adjacent polygon integration, the inconsistency of shared boundaries is a common question.Neighbor polygon overlay is not able to strictly guarantee topological consistent but generate series of fragment area and gap area in overlap region.For polygon map data, such as land-use, vegetation, and soil class, the boundary should be consistent and all polygons should cover the whole region without gap or overlap area.Shared boundary correction is also a necessary operation in polygon map generalization, representing as conversion from visual neighbor relation to topological neighbor relation between polygons.From this perspective, spatial relation generalization is one of important contents in map generalization.To correct the inconsistency between polygon boundaries, this paper offers different solutions based on spatial neighbor relation analysis.Three cases of inconsistency are distinguished: intersecting, separating and interlacing. Intersecting inconsistency causes small fragment area and separating inconsistency causes short gap area. Correction of inconsistent shared boundary is to find single line which can approximate two original boundaries with high location precise. Considering the different importance of two neighbor polygons locating beside corrected boundary, we provide two correction approaches: less important polygon to snap its neighbor, and equal polygons evenly to adjust their shared boundary.So 6 correction methods need to be separated: ①intersecting + snap;② intersecting + even;③separating + snap;④separating + even;⑤interlacing + snap;⑥interlacing +even.To find a method to separate inconsistent problem space equally, this study implements Delaunay triangulation model, a powerful tool in spatial neighbor assess, to detect inconsistent area among shared boundary, and further uses triangulation skeleton to produce common boundary. Through triangle stripping operation, the inconsistent area can be extracted as set of triangles, like a sausage.The sausage area is a simple polygon enclosing inconsistent area well.Skeleton construction based on sausage area has properties of simulating common boundaries well. Problem space is evenly separated by skeleton axis with the boundary location error adjust equally. Directly constructing skeleton based on Delaunay triangulation usually generates lots of short hair lines. In order to access single skeleton axis line, the study discusses the reason of producing hair branches and based on the triangle class characteristics presents one method of removing invalid triangle in sausage area.The presented algorithms have been realized in an interactive cartographic generalization software AutoMap which is developed by authors, and the running correction hardcopy result is illustrated in this paper.