The "correct" or perhaps "most correct" impulse is a Dirac delta so-called "function." Practically speaking, we cannot generate one. Steve Harris said that the impulse does not matter too much as long as it's wide band. Loosely speaking, and theoretically speaking, this is quite correct. But because we cannot actually physically generate Dirac delta "functions," it turns out to matter. You can see (hear) this for yourself by creating square waves, ramps, truncated sin waves, etc. and generating the response to these. One could argue that these don't qualify as impulses, but that's what we actually generate in any empirical situation --- a nonideal impulse. This is why I use calculated IR's rather then recorded ones (see below). The IR's calculated use as close to an ideal Dirac delta "function" as is possible on a particular machine. How much the impulse matters also depends on things further downstream, which are usually ignored in theoretical discussions. ---------------------------- Some comments were posted about specific IR's at some web locations, and I might as well address that here: I also really don't like any of the IR's that I've tried to use from various web sites to compare with those that I've calculated. The ones I've tried so far don't really sound any better than the DSP-oriented ones in my hardware synths. Some of them don't even sound as good. Wolfgang has posted that he uses convolution for everything except reverb. I use convolution with the IR's I've calculated for reverb, echo, stereo separation, and equalization. With calculated IR's the stereo separation is so good that it mimics binaural recordings. I'm not at all convinced that using stereo mics in a cathedral to record an impulse would do as well because of the spatial extent of the mics as well as that of the source, but perhaps it can be done. This spatial extent limitation plus other practical problems with attempting to physically record an IR may limit the empirical approach.