This is rather simple if you have a large number of these precision caps, best from many different sources. The idea behind is, that if you have a large number of these caps, then the cap values will follow the Gaussian distribution curve and you can find a reference cap, by a suited methode:

Build the following cirucit:



Choose Ra = 2k2, Rb = 1k8 +1k_cermet_trimmer, C2 = 10nF/foil and take a CMOS555.
Take the best parts you can find, means at least 1% metal film resistors and multi-turn cermet trimmer (best from Burns).

Now assume that you want to measure 47nF FKP2 caps from WIMA, which you can buy with +-1% tolerance. If you insert an ideal cap without manufacturing tolerances, then the above circuit, properly calibrated of course, will deliver a frequency of exactly 4700Hz, representing a capacitance of 47.00nF. Only thing to do is to measure the frequency at output of ICM7555 by a suited frequency meter. Often, multimeters contain such a frequency meter.

How to calibrate this circuit?
Assume you have a large number of these caps. Then measure each capacitance with above circuit and notate the reading of frequency meter. It doesn't matter, that the circuit isn't still calibrated. Important is only that you don't change the setting of cermet trimmer during this procedure.
Sort all the caps and notate the readings, so that you can find each of the caps later.
After a while you will hopefully see that all the values differ by up to +-1%.

If you have a very large number, then you can take the average of all the readings. Now, choose one of the caps in your list, which's reading is equal to the taken average. This cap is then your reference cap. Normally, you will have not only one cap which hits the average, but a certain number of them. Put them into a box and handle them carefully. All these caps are reference caps.

Now take one of these reference caps, put it into the circuit (sockets for fast changes prefered) and calibrate the cermet trimmer for a reading of exactly 4700Hz.

If you don't have a very large number of these caps, then you can look for the highest and lowest reading of the list. Taking the average of both will allow you to find the reference cap then. Of course, with this methode you will probably not find the exact reference value, because the number of caps might be too little. So, try to buy a real large number of these caps (from many different sources!), so that the center of the according Gaussian distribution curve will hit the ideal value.

Again, this methode makes only sense if you have a large number of these caps. Also, they must come from many different sources! Don't take caps from the same production lot. These don't differ enough.
With a bit luck you can find a reference cap with an error of less than about +-0.2%. When using a very large number then the error becomes smaller and smaller.

If you start a new measurement, then always take one of your reference caps and calibrate the circuit first. Afterwards you can measure.
Take care, caps show a certain temperature drift. So, don't touch them too long with your fingers. If the measurement lasts longer, then recalibrate the circuit by the help of one of your reference caps (also the ICM7555 shows a temperature drift!).

After some months you will find, that your reference caps which earlier showed same reading tend to differ a bit. This is caused by long term drift, which is up to +-0.3% in two years at 40°C.

Also, your circuit should be well shielded against hum.

Kai