[nwallin]

Astronomers could reasonably well measure angles between objects in the night sky, and with some basic geometry, you can measure the relative distances between objects reasonably accurately. For instance, if you have a triangle ABC, and you know the angle ABC is 45 degrees, and the angle ACB is also 45 degrees, you know that the distance AB will be sqrt(2)/2 times the distance BC. If you have dozens of other points you want to know about, you can calculate the distances relative to BC as well. But what if you haven't the foggiest idea how many toises long BC is? (this was well before the metric system; toises was the unit of choice for Cassini) You either give units of toises for, for instance, the size of the Mars orbit with error bars of +/- 80%, or you give the size of the Mars orbit in terms of multiples of BC with error bars of +/- 5%.

Measuring the AU is fraught with errors of all sorts. For centuries it mostly consisted of exploiting tiny parallaxes on the Earth's surface between planetary bodies- for instance, Cassini and Richtie measured the parallax of Mars between Paris and French Guiana. But a small error propagates to a much, much larger error in the final result than relative distances between planetary bodies in AU distances. If your measurement of the parallax of Mars is off by one arcminute, your measurement is totally useless, but if your measurement of the angle to Mars is off by one arcminute, your distance to Mars in AUs is off by a few percent.

It wasn't until the 1960s when the JPL measured distances to Venus and Mars using radar that we were confident we had a good grasp on how long an AU was. But by that point, we had already measured the relative distances between the bodies in the solar system using the AU ruler relatively accurately for centuries.