There are a number of things that could be wrong, like the polygon crossing over itself (a figure eight) or perhaps the starting and ending points are not identical. We have seen tiny little sliver artifacts where the outside crossed over itself only when zoomed in really, really, far. ESRI products seem to like to make these little tiny slivers when processed in certain ways which become invalid when moved from shapefiles to SQL Server. ThinkGeo might have implemented more checks of their own.
Glad to hear that your problem has been solved Bob, and more, thank you for your help on this post Allen, you are right, we are using the CLR Types as our default geometry topology function which is more strict than NTS we used before, so maybe some geometries could pass the validation before we integrate the CLR types and now are regarded as invalid. If some one wants to work with the invalid geometries which are valid before, one property should be set: BaseShape.GeometryLibrary = GeometryLibrary.Managed; // use the old NTS logic. BaseShape.GeometryLibrary = GeometryLibrary.Unmanaged; // use the new CLR Types logic.