Currently, I have a spreadsheet in which I need to calculate the
cumulative distribution function of the standard z, as well the inverse
of the cdf for p(x).

For the sake of speed(?), I am using Application.NormSDist and
Application.NormSInv.

NormSDist is respectable, but the accuracy of NormSInv is laughable. I
was thinking of substituting both functions with custom methods.

For NormSDist() I was thinking of using the algorithm in Numerical
Recipes, and for NormSInv() I was thinking of using P.J. Acklam's algorithm.

The function that I am evaluating is of the form: (all variables are double)
y = NormSDist( (NormSInv(x) - a * m) / Sqr (1 - a * a) )

and I would be evaluating this function approximately 6000 times.

Would the substitution of the canned functions with custom ones
hamstring my project?