Short answer
Record at 24-bit. It leaves plenty of dynamic range and room for error. Use 16-bit for CD masters and most final distribution. Consider 32-bit float when levels are hard to predict, such as field recording or sudden loud sources. 64-bit float is a processing format inside your DAW, and most home recordings don’t need to think about it.
What bit depth controls
Bit depth is the number of bits used to store each sample. It sets how finely the level can be measured. That determines two things: the dynamic range (the gap between the quietest sound the file can hold and the loudest) and the noise floor.
Each bit adds about 6 dB of dynamic range:
- 16-bit has 65,536 levels and about 96 dB of dynamic range. It is the CD standard.
- 24-bit has about 16.7 million levels and about 144 dB in theory. Real converters reach roughly 110 to 125 dB.
Bit depth doesn’t change the frequency range. Sample rate sets that. Bit depth sets how far the noise sits below your signal. For the relationship between the two, see sample rate vs bitrate.
16-bit vs 24-bit
For listening, 16-bit’s roughly 96 dB covers the range of most playback environments. For recording, 24-bit matters more. It leaves a large margin below your peaks, so you can keep levels well under clipping without giving up a clean noise floor.
The practical rule is to record at 24-bit, and reduce to 16-bit only once, at the final export, with dither. See what dither is and when to use it.
32-bit float: what it does
32-bit float stores each sample as a floating-point number, with 1 sign bit, 8 exponent bits and 23 stored mantissa bits. Because of the exponent, the file can hold values far above 0 dBFS, the digital full-scale point. Its dynamic range is roughly 1,500 dB, which is effectively unlimited for practical use.
In practice, a 32-bit float file keeps a loud peak instead of cutting it off. You can lower the gain in editing and recover a signal that would have clipped in a fixed-point file. That’s useful for field recording, unpredictable live sources and internal DAW processing. FL Studio’s mixer, for instance, runs in 32-bit float, which is why a red track meter there doesn’t distort on its own; see why audio clips in FL Studio.
A 32-bit float file is about 1.33 times the size of a 24-bit file at the same sample rate and channel count. Several portable field recorders offer 32-bit float recording.
The catch: float doesn’t stop clipping at the converter
A 32-bit float file can hold values above 0 dBFS. Your audio interface may not. Most interfaces still use 24-bit converters that clip at 0 dBFS, and the analog front end can overload before the signal reaches the file at all. If the converter clips, the damage is already in the file.
So 32-bit float is a safety net for overs that get past your gain staging. It isn’t a reason to record hot. For how to set levels, see what peak level to record at.
64-bit float: a processing format
64-bit float is double precision. Its extra precision matters for internal processing, such as summing many tracks or running long plugin chains. No microphone preamp or converter records 64-bit float. Most DAWs process internally at 32-bit or 64-bit float, whatever the depth of your files, and some let you turn on 64-bit float processing or 64-bit summing.
The audible benefit over 32-bit float processing is small in typical mixes, and it can add some CPU and memory load. It is also unrelated to the 64-bit processor in your computer. For the full answer, see is there a 64-bit float?
File size and choosing settings
At the same sample rate and channel count, file size scales with bit depth.
| Bit depth | Approximate dynamic range | Typical use | Relative file size |
|---|---|---|---|
| 16-bit | about 96 dB | CD masters, final distribution | 1× |
| 24-bit | about 144 dB in theory | Recording and mixing | 1.5× |
| 32-bit float | about 1,500 dB | Field recording, unpredictable sources | 2× |
Reduce the bit depth only once, at the final export, usually as the last step of mastering.
Common misconceptions
- “Bit depth affects high-frequency detail.” It sets the noise floor. Frequency range comes from sample rate.
- “32-bit float makes recording impossible to clip.” The file won’t clip, but the interface converter still can.
- “24-bit always sounds better than 16-bit to listeners.” At normal listening levels, 16-bit’s dynamic range already covers most playback. The difference shows up mostly in recording and mixing.
- “64-bit float is a recording format.” It’s a processing precision.