I'm going to take slight issue with what Roy wrote about measurement
using a directional coupler. Though it's true that the receivers could
be considered to be measuring voltage, they are measuring it across a
well known load (very close to 50 ohms). When we set up a system to
test the performance of units we build, it's done very carefully to
calibrate them for input _power_, not voltage. The usual directional
coupler in a VNA is indeed a resistive bridge, with some care taken to
make the known legs and the source driving it and the load observing
the imbalance all 50 ohms (in a 50 ohm system). Generally a bridge
whose impedance is near the impedance you want to measure is a good
choice for a starting point for making accurate measurements, but it's
for sure not easy to build a 10k ohm bridge at 10GHz, for example...

I certainly agree with Roy that an understanding of the measurement
system and the error sources is critical to getting consistently good
measurements. In fact, I have several old ap notes from HP about such
measurements; those ap notes may still be available through the Agilent
web site, but I couldn't guarantee it. NIST (the US National Institute
of Standards and Technology) has done a lot of work in characterizing
errors in measurement, and they are also a source of good (but usually
mathematically deeper) info. Rohde und Schwarz is another source of
VNA and impedance measurement ap notes. It's such a complex subject
that I really wouldn't want to even try to get into it here. And why,
since it's already been dealt with nicely in freely available pubs?

Cheers,
Tom