Anderson-darling goodnes of fit test

[Sarah Perkins]

to whom it may concern,

I am currently working on a master's project which requires me to
compare observed data with modelled. To do this, so far, i have been
using the kstest2 function in matlab (version 7.0). At present, I have
not fitted any distribution to the data, as essentially, I am comparing
the modelled to the observed; that is, I am using kstest2 to compare the
EDFs to see if it is liekly that they have been drawn from the same
population. Upon scouring the internet, I came across a script written
by Paul Kienzle that uses the Anderson-darling test to compare the
distribution of a population to a known distribution (EG Normal,
unifrom,weibull,exponential). I have since emailled him with the
following question, but he has reffered me to this help desk as his
statistics are not quite up to scratch:

Is it at all possible to write/modify the Anderson-darling test to be
in the same format as kstest2? IE, is it possible to use this test to
compare two undefined PDFs with each other, rather than comparing them
to a known distribution?

The reason why I am asking is because my resarch into the literature
suggests it is not possible; the test statistic contains natural logs
and cannot be used for 0 values. Also, the formulae written for testing
a certian distribution with the anderson-darling test only exists for 4
distributions. Hence, any adivce as to weather the Anderson-darling test
can be used like the kstest2 function in matlab is most welcome.

Looking forward to hearing from you soon,

Sarah Perkins