When I have a PDF and want an epub, first I convert the PDF to HTML, which can be done with Adobe's free online converter at
http://www.adobe.com/products/acroba...linetools.html
Then I bring the HTML into Calibre and convert it to epub.
There will still be errors, but many fewer than if I bring the PDF into Calibre and go straight to epub, and it's fairly easy to fix the remaining errors while it's in HTML format--it usually for the most part requires a little careful find-and-replace-ing.
Caveat: I only work with novels and basic nonfiction (no columns, tables, etc). I have no idea how well my system would manage those things. Bottom line is that PDF is just not an ideal format to convert from, and the more complex the layput, the more problems you'll have in the conversion.