When you say you need accurate timing does that mean the intervals between your 20 Hz times have to be accurate or are do you mean the absolute time of the measurement (day, hour, min, sec) has to accurate? These are two different issues and need to be addresses separately.
Address the interval timing issue:
You say you want the error/uncertainty small, you should do the analysis to see how small this has to be first and post this metric. Maybe a standard external crystal will be accurate enough to meet your needs or maybe you need a temperature compensated oscillator or something more accurate. Without knowledge of how accurate it has to be you can spend a lot of time and money coming up with a result that is two orders of magnitude better than what you needed. Possibly using the PPS signal from the GPS to zero the timer as a means of reducing drift would be enough.
The absolute timing issue:
In this case I assume you are concerned about making sure your measurements where the time stamps compare very well with the time stamp on measurements collected from another instrument located in a different location. This case can be helped dramatically by the PPS signal from the GPS, I have used this when making acoustic measurements over an RF networked group of sensors.
Hope this helps,