I was a little surprised by this, since publishing in Nature and Science is sometimes viewed as a "second tier" prospect in the social sciences:
An average social science article published in Nature (2000-2010) is more heavily cited than an average article in any social science journal, with a "citation impact" of 51 (see the Reuter's article here).
The second best place to get your social science article? Science (citation impact = 35).
Nature and Science don't publish many social science articles, 65 and 80 respectively over the last decade, but the articles they publish seem to do well. For reference, note the citation impact for top economics-only journals (Quarterly Journal of Economics is top in 2008 with citation impact = 5) and the top political science-only journals (Political Analysis is top in 2007 with citation impact = 2.5).
A standard critique of Reuter's citation impact measure is that it counts an article's citations over a fairly short window of time just following that article's publication (2 years). In the social sciences, articles may remain as unpublished working papers for several years, preventing many of their citations from being counted in Reuter's analysis. Is this long lag in publication timing driving Reuters' finding? Probably not. Reuter's also publishes a measure of impact that spans a longer time window following an article's publication: the 28 years from 1981-2008. Giving an article in the Quarterly Journal of Economics almost thirty years to accumulate citations still leaves its average citation count (49) just behind a the two-year citation count for an article in Nature (51). Similarly, the long-run citation count for The American Political Science Review is 31, just behind the two-year citation count for Science (35).
[If you're interested in citations and the structure of human knowledge more generally, see this earlier post.]