1. Be sure hat you have a fundamental quartz and not an overtone type.

2. Be sure that Vcc starts-up within 5msec or so, better within 1msec. As Erik already said, every oscillator needs a kick to start. This can be a sharp Vcc turn-on or noise on Vcc or from the oscillator itself.

3. Usually these Pierce oscillators work well, if a high Q quartz is used. These quartzes usually have a very low equivalent series resistance and you will find them under the bigger quartzes! So, if you have a tiny SMD quartz the equivalent series resistance can be too high.

4. Each oscillator needs a stable phase shift as shown here http://www.8052.com/forum/read/79811 . So, adding a small series resistance at output of oscillator, as shown in the datasheet of your micro can add a bit phase shift to make the oscillator start reliably. But take care, this resistor will reduce the quartz drive level and can make your quartz start less reliably at higher ambient temperatures, where the tranconductance (something like a gain) of MOSFETs of built-in oscillator decreases. A good methode is to increase the series resistor in steps to find the maximum series resistor where the oscillator is still starting and then to take the third of this value. But this can be a long journey...

5. Quartzes behave very strange if the quartz drive level is exceeded. A good fat crystal can withstand 1mW, but todays SMD quartzes do only 100µW. So, changing to a bigger quartz and/or adding this series resistor can help.

6. Experiment with unsymmetrical burden caps like shown here http://ecee.colorado.edu/~mcclurel/iap155.pdf . This can increase the amplitude at oscillator input.

Take care, Anthony, each modification must be tested with dozens of boards under all ambient temperatures, Vcc rise-times etc. May be it's better to use a ready made crystal oscillator then?

Kai