The sun-planet mesh is factorising if the number of teeth on the sun is divisible by the number of planets.
The planet-annulus mesh is factorising if the number of teeth on the annulus is divisible by the number of planets.
In a factorising combination, the fluctuations in the gear mesh forces are all in phase giving high torsional force fluctuation but low lateral force fluctuation.
The AGMA documentation and MikeyP's summary pretty much describe the basic situation. However, while the basic differences that exist between factorizing and non-factorizing epicyclics may appear to be fairly straightforward in theory, in practice the situation can often be much more complex.
Depending upon the number of planets, load sharing among the meshes, the accuracy of the gears and their mounting structures, the pitch line velocity of the gear meshes, the relative meshing/passing/structural frequencies of the gears/bearings/structures, etc., either a factorizing or non-factorizing configuration can be the better choice.
Unless you have substantial engineering resources at your disposal to conduct a detailed analytical trade study to determine which approach is best for your application (factorizing/non-factorizing), then the safest approach would be to select gear tooth counts/geometry combinations for your epicyclic that give good load sharing, ease of manufacture, and have compatible meshing frequencies.
Sorry if I was confusing. And I hope you'll correct me if I'm wrong.
I meant "compatible" as in not tending to couple with other modes occurring in the drivetrain. For example, if your epicyclic was the output stage for a propellor speed reduction gearbox, you probably would not want to have a fundamental mesh frequency that was a multiple of the propellor blade passing frequency.
I believe there is a table in AGMA 6123-B06 that gives fundamental meshing frequencies for various epicyclic configurations & tooth counts.