If one were to analyse a large number of chess games, which specific piece (not type of piece) would be found to be taken the least number of times? And yes, in light of SomePatzer's comment, besides the kings of course.
I recently looked into that very thing a couple of weeks ago, and found an excellent breakdown on it (just this one, though):
The short answer is that aside from Kings, h2 & h7 are the least likely to be captured pieces (with survival rates claimed to be 73.92% and 72.29%, respectively).
Regarding individual pieces' survival (from a 2,196,968 game database), http://www.quora.com/What-are-the-chances-of-survival-of-individual-chess-pieces-in-average-games/answer/Oliver-Brennan gives a table (and an accompanying graphic) of each piece's respective chance of survival.
While I was at it, I found some data on square usage. I've included the sources below not because square usage relates to your question directly, but because the nature of the two topics is so close that I figured that they might be of interest to you (or future readers).
Using 509 games selected randomly [and analyzed] out of a 567,000 game database, these two URLs give tables of raw events and normalized through-traffic usage of squares, and heat maps of same:
This URL analyzes the moves of 12 Grandmasters' entire careers (separated by white and black, all normalized [no raw counts]), along with some graphs of average-data counts:
There are also some excellent analyses (per side) of each type-of-piece's square usage at http://philanalytics.blogspot.com/2014/11/chess-analytics-analyzing-championship.html .
Probably the rooks. That's why the greatest number of endings are the rook endings.
The knights and bishops fight it out early on and get "killed." Later, the queens are targets because they're so powerful, and they get traded. The rooks are left on the board. Sometimes one or both sets are traded, but often, they remain until one player resigns.
Between the rooks, the queen rook is less exposed, and therefore less likely to be captured.
The obvious answer is the king. It will be taken 0 times. The most frequent one would be a pawn: there are 16 pawns and you can hardly find a game, where a pawn is not taken. I would bet that the least frequently taken (apart from king) would be a queen.
I do not have statistics to back up my assertion, but you have only 2 queens and 4 rooks, bishops, knights. Also pro players prefer not to exchange queens early, so they will go only after some of the other pieces would be exchanged.
The h2/h7 pawns as confirmed by @Charles Rockafellor seems to be the most likely to me; at least for normal games above Patzer level.
If you look at games there are very few pawnless end positions. And since more people castle short than long, it makes sense to keep those protective f-g-h pawns a bit longer. Also the h pawns being where they are at the rim, it is very hard to get at them and if they'd ever be involved in some exchange action it could only be to one side, towards the g file.