I'm not really happy with the question, and the explanation is just wrong.

When I saw the question, I knew it was going to be related to inaccuracy in the floating point represtentation. So when I saw the answers, I immediately knew to choose the one that made the least logical sense - that's always the case with these kind of questions. Luckily, there was only one illogical answer. (If "3.7/3.7" had been offered as well, I would have had no choice but to run the query, as these answers are impossible to predict unless you are willing to manually convert it to a binary floating point representation in 53 (I think) bits.

I like the answer even less. There are no tie-breaking rules used. If there were, then the answer should have included a reference to a page where these rules are actually documented, and this is defintely not the case. The specific answers returned by this single query may seem to suggest that STR() uses "bankers rounding", but that really is a mere coincidence. Many other values will round differently, as several posters in this topic already have explained.

(By the way, I did some research and found that there has been a discussion on perceived bankers rounding by STR on this very forum - check http://www.sqlservercentral.com/Forums/Topic246556-8-1.aspx).