The -72db figure that you are seeing is accurate, but a bit misleading.
Audacity works internally with 32-bit float data. This offers an extreme level of precision to processing tasks and the theoretical maximum SNR (signal to noise ratio) is some ridiculously high figure (very good).
Unfortunately, few other applications support 32-bit float format, for example CDs (and many other applications) use 16-bit (integer). At some point it becomes necessary to convert from Audacity's high resolution format to a much lower resolution. Ideally this conversion should be done only once at the very end of editing and processing of the audio.
The dynamic range of 16-bit audio is generally quoted as around 96 dB. This is because each "bit" provides a resolution of about 6 dB, so for 16 bit the amplitude scale goes from 0 dB down to -96 dB. When you get down to the last few bits, (below about -70 dB) the sound quality becomes terrible because the waveform is just jumping between a few binary values.
What Audacity does (and most other high quality audio processing applications also do) is to apply "dither" when converting from a high resolution format to a lower bit depth.
Dither of audio is similar to the use of dither in graphics which can smooth the appearance of text and images by smoothing the steppiness of edges. Perhaps surprisingly this can actually increase the dynamic range of 16 bit audio. This is because, although dither adds some broadband noise, it becomes possible to reproduce sounds that are well below the noise floor. (see here for a bit more explanation
http://wiki.hydrogenaudio.org/index.php ... requencies )
The -72 dB noise level that you are seeing is a peak level measurement and represents the maximum noise deviation of +/- 2 bits of "shaped" dither. A more meaningful measurement would be to look at the RMS level (as this relates more closely with how we hear). The RMS level of shaped dither is -82 dB and the majority of this noise is at very high frequencies where human hearing is least sensitive at very low amplitude. If your recording has a maximum level close to 0 dB, the dither noise should be virtually inaudible unless you have the volume turned up to ear-bleeding levels.
If you have a specific need to absolute silence in a recording, there are two options:
The best sound quality option is to keep everything in 32-bit float format.
The other option is to temporarily turn off dither is Audacity Preferences (Edit > Preferences > Quality) but be aware that this will reduce the sound quality of non-silent audio when processing or converting from 32-bit to 16-bit. Normally it is best to keep dither enabled for "High Quality Conversion" and disabled for "Real Time Conversion". These are the default settings in Audacity.