The researchers used the simple method of giving programs random streams of input (the so-called "fuzz test"). They found that while the commercial Unix systems had failure rates of over 18%, Linux had a failure rate of 7-9%. As the authors say:
"It is reasonable to ask why a globally scattered group of programmers, with no formal testing support or software engineering standards can produce code that is more reliable (at least, by our measure) than commercially produced code..."
The paper is avaliable at http://www.cs.wisc.edu:80/Dienst/UI/2.0/Describe/ncstrl.uwmadison/CS-TR-95-1268 .
A recent study by Bloor's also found that Linux was quite a bit more stable than Windows NT. Over the course of a year, Linux was down once, for four hours. During the same year, on identical hardware, Windows NT went down many more times, and was down a total of 65 hours, or over sixteen times as much.
In a 1995 interview with Bill Gates, the German magazine FOCUS appears to have uncovered an interesting attitude towards bugs on the part of Microsoft's chairman. Whether his attitude has changed since then is an open question.