I'd try going back to the source of your jpg map and see if you can get output a pdf. The potential is that you can extract vector data from the original pdf using Adobe Acrobat, Photoshop or some small utility like "pdf2cad". The vector data will be in dxf or dwg format and can be georeferenced and converted to feature classes using Conversion Tools: CAD to Geodatabase. Only problem here is separating some of the features. Best is if you can output pdfs of exactly the same area but showing only one type of feature at a time. Makes for easier reconstruction as feature classes.
If you do not have any of that then to avoid the sliver polygons I would:
1. first trace all the dividing polylines and attribute them geologically (do not make polygon shapes).
2. Then I would make a new polyline feature class, select all the lines that outline polygons, right click copy, right click paste into the new feature class (Task: create new feature; Target: the new polyline FC) and MERGE the copied set. Make this FC selectable and the original not.
3. NOW in a polygon feature class create a new single polygon covering the area of the map.
4. Select the polygon and set the Task: Cut Polygon Features.
5. Click the Sketch tool (pencil in 9.3) and snap to the merged set of lines.
6. Right-click the line and hit Replace Sketch from the pull down. Vertices of the line show up.
7. click F2 or "finish sketch"
The merged set cuts the polygon into as many polygons as there are closed polylines and then you can attribute the individuals. The method is described in Desktop help, but only for a single line cutting a polygon.
Then overlay the original line feature class with its respective geological attributes and you have a map.
There are a couple of geoprocessing tools to either highlight slivers (thiness ratios) or to integrate two closely adjacent features like slivered edges of polygons or a polygon and polyline that are supposed to represent the same boundary, but I have not had success with them and avoid that route.
Hope one of those methods works for you.
Hardolph
