I'm currently moving through the literature on RNase III. I'm in 1997 now (for my current work, it always pays to go older to newer, so I've been following the literature from the early work in the 60s toward the present). Up to 1997, very specific cleavage by RNase III had been demonstrated without identifying any conserved motifs that appeared in the cleavage sites. Zhang and Nicholson had the bright idea of comparing all the confirmed cleavage sites and figuring out what wasn't there, and in so doing turned up unfavorable base pairs that disrupt RNase III binding, preventing cutting at those sites. Thus, specificity defined by a lack of certain features.
I love papers like this.
The Pubmed entry for the paper. You can read the paper for free at PNaS or Pubmed Central, both of which are linked out from the Pubmed entry.