Consideration of when Right Association works and when it fails lead to a restatement of this parsing principle in terms of the notion of heaviness. A computational investigation of a syntactically annotated corpus provides evidence for this proposal and suggest circumstances when RA is likely to make correct attachment predictions.