Here's the simplest procedure to get realistic-sounding IRs:
You need enough floor space in the test area to place the speaker and mic 2 meters apart, with no reflections that will arrive less than ~20 ms after the direct sound from the speaker. Practically speaking, this requires a room that is ~8.1M (26 1/2') x 6.1M (20') or larger. The speaker and the mic should be placed near the center of the floor space so as to maximize the reflection-free zone.
Place the mic on the floor. Ideally, it should be a hard surface. If you have carpeting, use large sheets of cardboard to make a reflective surface that surrounds the mic and speaker, which can also be on the floor. I usually take IRs at 4 angles: on axis, 15, 30, and 45 degrees off axis. You can easily mark positions on the floor (or cardboard) so you can quickly move the mic.
With the mic on the floor aimed at the speaker, the reflection from the floor that will also be part of the measurement arrives so soon after the direct sound that its only effect is to add 6dB to the level the mic sees. Since there will be floor reflections added by a player's rig, this is the best way to get the sound of the cab you're measuring. However, there's a trick I use to get a "bonus" cab: place the mic off axis of the speaker, about a meter off the floor, and the floor reflection adds the effect of an virtual "image" cabinet that is located under the real one, below the floor. In this way, you can use a 2x12 cab to get the IR of a 4x12 with the same speakers in it, and a 4x12 can produce the IR of a full stack.
The test signal that will yield the best results under less than ideal conditions (e.g., nonzero environmental noise) in the shortest time is a swept sine. Specifically, I recommend a log sweep for IR acquisition. Send the signal to the speaker at a healthy level, and record the mic output. Then use a deconvolution routine to process the recorded sweep to produce the IR.