How to do it

Let's say you're a creationist, and you've decided to make a lot of noise about evolution. So you reach for your keyboard and start typing:

"The human eye is composed of so many different interlocking parts that it can't possibly have..."

STOP RIGHT THERE! You've just fallen for a schoolboy error: you're about to make an argument for which thorough, accurate and catchy refutations are available. Probably you read this argument in an ISCID pamphlet, right? Didn't you consider that evilutionists may also have read that pamphlets, and prepared themselves to rebut its claims?

Instead, and I cannot emphasise this enough, you should pick an argument that they haven't come across before. Dembski actually had the more mathematically-inclined evilutionists feeling uncertain for a bit. Behe managed to look convincing for at least half an hour. I repeat: the best arguments refer to areas of academia the evilutionist is unlikely to have hitherto explored.

Exhibit A:

"It's simple, really, but the best ideas always are. Make a graph whose vertices are all possible genotypes with two vertices connected if they are one mutational step away from each other. That graph is isomorphic to a Cayley graph of a certain matrix group with respect to a standard generating set. (Surely that's obvious?) Such Cayley graphs attach in a natural way to arithmetic Riemann surfaces, as I explained in obnoxious detial in Chapter Five of my thesis. It is now a consequence of Selberg's eigenvalue conjecture for such surfaces (which everyone just knows is true) that these graphs have weak expansion properties. That is, they have relatively small Cheeger constants, which implies that they fracture easily. Which in turn implies that evolution by natural selection can not move efficiently through the graph. QED."

Pure genius at work!