A couple of notes there: first the item at the top of the list has more than an order of magnitude more ratings than the rest. For most ordered lists, this is probably an edge case that may have to be dealt with separately.

Finally, I've got to say, 26 reviews (for the #2 item) doesn't seem insignificant, and the #2 item seems to have something like a 15% higher _average_ rating. Also, on another trending rating (http://www.rateitall.com/t-3239938-2010-ncaa-tournament-team...), the #2 has a significantly higher rating (17-18% it looks like), but only one less review: 4 vs. 3. I think if I were sorting either of these based on the average rating and the # of ratings, I would have done both differently.

Based on this, it seems that the Wilson score probably _over-emphasizes_ sample size, especially on things like ratings on high-traffic internet sites that may have orders of magnitude swings in the number of ratings.

It seems like the Amazon method of average works just fine, except for items with very few ratings possibly receiving disproportionately high ratings. Again, edge cases, which should probably just be penalized manually.