I've never found a compelling argument either way. If the level is really lwo to start with I tend to normalise it first or at least apply enough gain so I can hear and see what I'm doing.
However I wouldn't normalise in Vegas, SF has a Normalise tool that's probably better than Vegas's, not by much if any but it's there so why not use it.
However using Normalise mighn't be such a wise move, use the SF Scan in the Normalise tool or Statistics to findout how much gain you can add and then use the Volume tool at add 3dB less or else set the normalise to -3dB. The notch filters in NR2 might introduce small amounts of gain at the corner frequencies so safer to leave a bit more headroom.

Bob.

If I have to normalize I do it first. If you need do it a lot then aim at getting better recording levels in the first place because it does introduce noise. If your levels are near to correct don't normalize just increase the volume envolope. I think normalization is for solving problem recordings only and not a natural thing to always do.

Doing a scan in SF before normalizing gives a reading of the noise floor. Careful setting of a noise gate can be more successful than noise reduction when the noise level is below the recorded material level. Most times, the noise will intrude into the recorded material, but, setting a noise gate gets rid of the noise below the gate setting.

Just addressing what may be a few misconceptions in this thread so far ...

Normalizing or adjusting volume does NOT introduce noise. Any noise you hear after doing these things was already in the recording. The noise level increases the same amount as the signal level so it sounds like there is more noise afterwards, but there is just as much more signal too. There isn't any new noise.

Noise gate does not remove noise, despite it's name. It removes all signal whenever the total signal is below a certain level. If you set the level too high then noise gate will happily blank out speaking, singing, and music too, should that drop below the gate level. And, when the noise gate passes signal it's also passing all the noise too. The reason it seems effective is that in decent recordings the noise usually is the quietest component and it's most noticeable when other sounds aren't present. The noise gate will then blank out everything during these quiet periods which is usually when the only thing present is noise. When the signal is strong enough to pass the gate, the noise is still there, but hopefully drowned out by the speaking/music/whatever it is that you want to hear.

The compelling reason to run noise reduction first is that noise reduction is only effective on a constant noise. If do any other processing first, especially something that changes the dynamics such as compression, RMS normalization, or wave hammer, the noise is no longer consistant. No matter where you choose the noise sample, i'ts pretty much guaranteed not to match the noise pattern anywhere else in the file. Even a straight volume increase can cause problems due to quantization. Noise will be increased differently depending on what other signal it's mixed with and will again end up with an inconsistent pattern.

For that matter, if you've adjusted the volume while recording, you really should split up the recording into each section with a different recording level and apply noise reduction to each section individually.

Peak normalization and volume increase, either through a fader or a volume envelope, are the same thing. The only difference is that peak normalization increases the volume based on an automatic comparison of the loudest peak and the desired level. Other volume increases are a fixed level of the user's choosing. If, for example, normalization decided it needed to increase the level 14.3 dB, then doing a volume increase of 14.3 dB would give exactly the same result. Therefore, normalization is a completely ordinary thing to do.

RMS normalization is a slightly different beast in that it decides the volume increase amount variably from moment to moment along the recording instead of using only the single loudest peak.

The only time i would recommend a volume increase before noise reduction is if everything is so low that it's nearly inaudible. In these cases you can suffer quantization errors because value of the samples is so low that the calculations involved will suffer rounding errors that will make up a substantial amount of the level of the recording. However, in these cases you'll be far better off re-recording the material at a higher level because there's probably too much noise and too little resolution to produce a worthwhile recording anyway.

I think I agree with most of that, the noise gate that ships with Vegas is pretty woeful. The Graphic Dynamics from SF is much more useful with careful tweaking, I've used it quite a bit on old recordings, particularly when they're being encoded to mp3.

RMS Normalisation in SF isn't dynamic from what I can see. The tool scans the region and derives an average RMS value that it then uses to apply the rquired gain to achieve the set RMS level. This can easily cause clipping unless you also use dynamic compression. A quick and dirty way to get all your tracks sounding the same volume.

NR2 is only a multiband notch filter, I usually tweak the noiseprint before applying the filter but this can be difficult to do if the recording level is low as the noiseprint window has a linear scale from memory. Even if you were to apply compression before using NR2 (not that there's any reason to do so) it's very unlikely to affect the noise spectrum, the noise would be below the threshold.

One trick with NR2 is getting a good sample and the duration of the sample needs to include a number of cycles of the offending frequency, using a very short sample window to get a noiseprint to remove a 60Hz hum just doesn't work. Use a long window and you risk getting frequencies in the sample that you mightn't want attenuated. Manually tweaking the noiseprint is one solution or else creating a pure 60Hz tone in SF and taking a noiseprint of that and then applying that noiseprint to the recording is another solution I've found.

Absolutely! A noise gate is incredibly destructive if you set it high enough to blank recorded material. Yet it will work miracles if you use it carefully. I alway preview the setting and adjust to taste.

As you point out, raising the gain on a piece, also raises the gain on the noise. Skilled use of a dynamic multiband compressor can hep..but again, very carefully! What you really want to do is raise the gain of the peaks, without going over 0dB and not boost the noise floor at all.

I stand corrected in the "pedantic sense" that normalization only expands the wave form and as such amplifies all audio components by the same amount - but the noise is still amplified! With proper recording the noise will be minimized. The big point is that pros who record right just don't need to normalize.

That's an interesting comment. I guess pro's never make mistakes.