NF Distance = D² / λ - a good rule to remember.
NF Distance is the distance from the speaker to the near-field/far-field transition point.
D is the diameter of the speaker (effective diameter; for a typical 6.5" woofer this is about 12.5 cm).
λ is the wavelength in question. Basically, in cm, it's
34300 / f where
f is the frequency in question.
When you are closer than this distance, you're in the near-field of the speaker. And the output of the speaker is "chaotic" in that it does not follow the typical 1/r² fall-off, nor is it even continuous.
For a 6.5" woofer tested at 1900 Hz, the NF Distance would be (12.5 * 12.5 / (34300 / 1900)) 8.6 cm, around 3". Distances closer than that are in the near-field; distances beyond that are in the far-field. Frequencies higher than 1900 Hz would have a larger NF Distance; frequencies lower then 1900 Hz would have a smaller NF Distance.
Note that this equation basically means the near-field INCREASES with increasing frequency. If you're making frequency-amplitude-dependent measurements within this distance you are not getting an accurate measurement of the speaker's performance in a typical installation.
Also note that it implies bad things for higher harmonics when doing distortion measurements. You can significantly skew your measurement results because the harmonics may be higher AND lower than what they would be in the far-field, where you usually sit. For the example above, a measurement distance of 10 cm would be about 2200 Hz; if you did a THD measurement at 1200 Hz, you would be comfortably outside the near-field for the fundamental, but ALL the harmonics would be in the near-field, meaning your THD measurement is highly suspect. And will change with just small movements of the mic.
Note, too, that when you reach a dip in the response because of the nearfield the impact of the noise floor is increased; you can see a THD contour that quite reasonably is the inverse of the frequency response at the harmonics.
Personally, I don't use near-field measurements if I can avoid them. I use them when troubleshooting a system (trying to track down an airleak, or incomplete glue joint), or when I have to because of the nature of the system (like headphones, or computer speakers). But otherwise, good 0.5m or further (I prefer 2 meter anechoic) is a good way to go.
For reference, I've been doing a TON of THD/IMD/SC (spectral contamination) and other non-linear effects (such as compression - power and BL) measurements the last year for one of my clients - Microsoft - in the development of some new telecom products (the
RoundTable being one such unit). Distortions - linear and nonlinear - are EXTREMELY detrimental to echo cancellers, and as such must be dropped to as low as possible (and yes, there is an XBL² speaker in that unit - it comfortably dropped THD by 60% compared to the best underhung units tested).
The first part was getting reliable THD/IMD/SC measurements. Thankfully we had a way of checking the results, because with proper modeling and hard-core theory, we could predict the level of echo cancellation for a given THD/IMD/SC level from the speaker. If something didn't line up, we could check the setup. It took a few months to get things dialed in so that it all agreed with physics, theory, models and measurements. This included construction of an anechoic chamber, tightly-calibrated ultra-low-distortion mics, and quiet buildings.
Oh, and it's a real two-way effect; your microphone also has a near-field associated with it. In most cases it's not an issue since most mics are small diaphragms, but with the larger 1" units you still have issues with getting too close for high frequency THD measurements.