You might want to check out my post on this subject in the General LC forum.

Even the water-based water density measurements - which are the accepted gold standard for %BF calculations - are only an estimate because of the very assumption you mention. You have to assume that the body is made up of just the two tissues and that you know the density of the two. That assumption imposes, as near as I can determine, about a 5% margin of error on BF measurements. This can be reduced by using frame structure and race-based corrections to compensate for the amount of bone and the different bone densities that are known to have a strong racial correlation.

The only other method, that I am aware of, that comes close is BIA when it is done according to very precise protocols. Even then it appears to be at about the 10% margin of error.

So as not to confuse people on the difference between a 10% error and a 10 percentage point error, if your measurement came out to be 20% fat, then a 10% margin of error would place your actual value between 18% and 22%.