How to use the LOGNORMALDIST function to generate a Cumulative Distribution Function? In Excel or Google Docs you can readily construct a Cumulative Distribution Function for the Normal distribution using the NORMDIST function (X, Mean, Standard Deviation).  I am trying to do the same for a Lognormal distribution using the LOGNORMALDIST function that has the same parameters (X, Mean, Standard Deviation).  I have played around with using X or LN(X) on the mentioned parameters. But, I get absurd results.
Do you know how to generate a Lognormal Cumulative Distribution Function using LOGNORMALDIST? 
Thanks in advance for any assistance.  I will readily reward anyone with a good answer with a positive vote.  
 A: On Microsoft Office for Mac 2008 LOGNORMDIST(5;1;1) gives 0,728882893 in Excel. On R plnorm(5,1,1) gives 0.7288829. In R you need to supply mean and standard deviation on log scale, so it seems that in Excel you need to do the same.
A: I mistrust all but the lowest-level functions in Excel, and for good reason: many procedures that go beyond simple arithmetic operations have flaws or errors and most of them are poorly documented.  This includes all the probability distribution functions.
Numerical flaws are inevitable due to limitations in floating point accuracy.  For example, no matter what platform you use, if it uses double precision then don't try to compute the tail probability of a standard normal distribution for z = 50: the value equals $10^{-545}$, which underflows.  However, using NORMSDIST you'll get increasingly bad values once |z| exceeds 3 or so, and once |z| exceeds 8 Excel gives up and just returns zero.  Here is a tabulation of some of the errors (compared to Mathematica's answers):
$$\eqalign{
 Z &\quad 10^6\text{(Excel - True)/True} \cr
 -3 &\quad 50 \cr
 -4 &\quad 468 \cr
 -5 &\quad 1580\cr
 -6 &\quad 3582\cr
 -7 &\quad 6462\cr
 -8 &\quad 70789\cr
 -9 &\quad -1000000
}$$
Therefore, if you must use Excel's statistical functions, severely limit the ones you do use; learn their flaws and foibles; work around those problems; and use the ones you are familiar with as building blocks for everything else.  In this spirit, I recommend computing lognormal probabilities in terms of NORMSDIST: just apply it to the log of the argument.  Specifically, in place of LOGNORMDIST(z, mu, sigma) use NORMSDIST((LN(z) - mu)/sigma).  Once you have verified that these expressions do return the same values, you have also established exactly what LOGNORMDIST does: there's little possibility of confusion.  In particular, you can see that mu is the mean of the logarithms, not the geometric mean, and that sigma is the SD of the logs, not the geometric SD.
