Deconvolution—I want it all!

A groupmate and I were discussing practical deconvolution of an instrument's response from data the other day. After some searching on the internet, he came across the following words of advice:

When it comes to deconvolution, don't be greedy.

This is actually pretty good advice. In theory, the idea works great. Fourier transform the data, divide by the instrument's transfer function, then inverse Fourier transform to get the deconvolved data. But with real data, you risk amplifying the noise or introducing artifacts, especially with FFT-based methods. This last point is pertinent if your instrument's impulse response is very short compared with the span of the data. So use some caution; just because you can deconvolve, that doesn't mean you should.