I don't see anything boldly wrong here, my thought was acquisition time.  I see
you use 20uS (verified?) and the book says 11 or 12.  I always used more.  Most
of
the time, I select a channel as the last act in the routine, then start the
conversion when I get back to the start of the routine next time around.  There
is
then no doubt that settling has taken place.  I also no longer use the AD
interrupt, mainly because the conversion time is so short, I just loop for the
result.  Just like you do below.

Technically, your reference voltage is below the spec.  Also, there is a maximum
signal source impedance that you must honor, and your 10K pot is questionable.
Many times it is best to buffer a signal so that it has low (<10ohms) impedance.
I would be careful to use only 5V rails, and no negative on the opamp.  It is
imperative that you do not go beyond the PIC supply with the analog value. (voic
e
of experience).

Last issue, I found that the electrolytic cap back at the supply does not offer
good enough filtering at the PIC (when seperated).  I put a 100u at the PIC and
reduced bobble dramatically.

Good luck in your quest.

Chris Eddy, PE
Pioneer Microsystems, Inc.

PIC development wrote:

> Has anyone who has used the example code for the A/D convertor
> on the 16c74A noticed "leakage" between channels?
>

etcetera, etcetera.