Measuring capacitance is hard because a lot of other things play a roll. For instance frequency behavior, parasitics, ESR, leakage, dielectric absorbtion, tempco, voltco, piezo electric effects, and then things like shielding and guarding. You need a very, very, very good cap and meter to measure the same value for a certain bandwidth.
Like Conrad Hofman I think a bridge circuit is the best option unless you can use some factory calibrated IC. But only if you build a correct circuit and measure ideal caps in an ideal fixture.
My best capacitance meter is a 0,01% GR-1620 bridge that is still in production (you really do not want to know the price but it is one of the best, if not the best, on the market. Measuring a cap to full accuracy takes something like 10 minutes and that a few times because it is very hard to do it right. Everything must be shielded and the zero detector used to the fullest sensitivity of 1uV for a full scale. I also have some capacitance standards.
I have a GR-1608 back then stated as 0,1% but it turned out to be much better and is later re-spec-ed, I had to replace a resistor, that was almost 40 euro. Not only the initial value accuracy is important but also tempco and things like sensitivity to humidity. This was the cheapest choice to get the right performance (it was advised to me by the original designer of the bridge who helped me with advice and documentation)
My two daily meters are 0,1%. To measure with 0,1% is a whole different story. Both uses a 5 terminal measurement. If I use them 2 terminal I will not even come close to 0,1%.
But one thing I missed in the discussion,
All meters that are capable of getting this kind of accuracy measure the complex impedance or admittance. You can not measure only capacitance if you want 0,1%. You need to measure the reactance (and calculate C from it or scale it in capacitance) and real resistance part too. Or only measure calibration grade caps with an ultra low loss angle. For pF you can use aircaps or vaccuum caps, for 100 pF and up Mica is a good choise but you need to calibrate them and select them for lowest DF, 100 nF and up I think polystyreen as dielectric.
I tried to make my own caps by calculating them but I not even came close thanks to things as the fringing effect. The dielectric is the biggest problem, If you only use air and two guarded rond plates it is doable but still hard, besides that it becomes huge (Marconi builded a big barn for them to hang from the ceiling) I made a coaxial one that came close (only 50% off
) but because the core would not float in air alone it needed some support and that made the fault over 100%.
I tried oscillators, F.E. Terman writes about a methode in his book about measurements. I tried that but it was hard to get it right. The oscillator was very hard to get stable enough (and you need to measure that with a very good counter) The biggest problem I could never figure out. As soon as I connected the CUT all went well and I could very precise measure the frequency shift but if I removed it, the oscillator had a higher frequency as before and would not go down unless I switched if off a while.
But without calibration grade gear you will start a very frustrating journey. For one because you will never know if you reach 0.1%
The easy way is matching them. Use a sinewave and measure voltage and current, oh wait, you only have a 3 digit meter (and low BW ?)
Then use two caps in series. Use an ac signal and measure the voltage over both. You can calculate the reactance of the unknown cap from the ratio and then recaculate the unknown capacitance. Or you build a bridge....(and even then forget the 0,1%)
Just use the best caps you have/buy or maybe there is someone here nearby who can measure them for you (although I think only a few of us really can measure a tracable 0,1% and then for a limited frequency range. Do an impedance measurement from your best cap from 100Hz to 1 MHz and you will not get a straight line.