>I ran it last night and the next morning it was late with about 1/4
>second.


I've just sat down with my calculator. It looks like  1/4 second in 12 hours
(43200 seconds) is just short of 6ppm deviation which could well be within
the tolerance of your crystal. I have a catalog in front of me which quotes
+/- 20ppm for their 32 kHz crystal. One ppm corresponds to 0.6 seconds a
week and even that is quite hard to achieve by adjusting the oscillator.

If you need real longterm accuracy you might have to use an eeprom location
to store corrections for each individual crystal. A correction of +/-16
seconds every ten days should allow you to get within 1ppm over a +/- 15ppm
range. You will also need an accurate clock to calibrate them against,
probably using a thermostated oscillator.

There are two furtber issues you will need to address. Firstly temperature
coefficient, and secondly ageing.  The 32kHz crystal I was looking at has
quite a small temperature correction, less than .05ppm/degree  This is
unusually good because the crystal is made for use in timing applications.
Ageing is another cause of errors, and one which is hard to quantify, look
_hard_ at the manufacturers figures for ageing because several ppm drift is
not uncommon.

I hope this is some help,

Keith.
==========================================================
Keith Dowsett         "Variables won't; constants aren't."

E-mail:      kdowsett@rpms.ac.uk
Phone:       0181-740-3162
Fax:         0181-743-3987

Snail mail:  MRC Clinical Sciences Centre, Cyclotron Unit.
                 Hammersmith Hospital. London W12 0NN.