Jerry,

I'm ignoring all p 0.5. Then I'm using

inv_tdist(p,df) if q < 0.5
inv_tidst(1-p,df) if q 0.5,

where q = rand() is an uniform r.v. from Excel (independent from the
previous one).

I don't why it's faster, but it is! At least 5 times faster, if not 10
times. I'm using Excel 2003.

But I think I found that my problem is in the statistical domain, not
in the computation. See this thread:
http://groups.google.com/group/sci.s...83e1dd956d59e/
or sci.stat.math thread with subject "t distribution -- sample
kurtosis" if this link doesn't work. (Anyone knows how to link
correctly to threads?..)

Btw, is inv_tdist in public domain?

Thank you!