IR spectra using a KBr disc or a diamond ATR. Which is better?

The old fashioned way to run IR spectra is by preparing a KBr disc. Sometime in the last decade, the diamond attenuated total reflectance (ATR) was introduced. Some people consider this input method to give the same result. However, the KBr disc method gives greater resolution than the ATR. You'll also find that the ATR method gives a lot of interference below 600 wavenumbers, which results in it only being practical to measure down to 650 cm-1, whereas with the KBr disc method one can go down to 450 cm-1.
Compare the spectra of the same compound from a recent publication of mine. I've analyzed the same compound using both methods so as to illustrate the difference. You can see that some of the smaller peaks that are clearly distinguishable in the KBr spectrum get lost in the noise in the ATR spectrum.

Spectrum taken using KBr disc method

Spectrum taken using ATR method

Note that both spectra were obtained with the same instrument and the same editing software.

So which should you use? If the compound is easy to make and its structure has already been proven with NMR and mass spec, then the IR is mostly a formality and you don't need the high resolution version. It might take you 20-30 min to prepare a KBr disc, while the ATR method will only take 5 min. The ATR derived spectrum still provides an unambiguous indicator of the unmistakable  functional groups such as carbonyl, hydroxyl and nitrile. However, if the compound is hard to make, you ought to characterize it in high resolution so that, especially the fingerprint region, can be matched at a later date (potentially in a natural product synthesis).
My advice is to use the ATR method when you need to characterize a library of compounds. If you're only analyzing a small handful of compounds (less than 10) then stick to the traditional KBr disc method.