You can actually get EBSD-like patterns with a flat sample, but the contrast is reversed.  You have to lower the working distance and point your EBSD camera downward, but it's a fun experiment.  I've mapped a few materials in an older JEOL 6400, where we could easily lower sample and position camera.

The history, as I've heard, is that 70 degrees comes from using (100) Silicon as a pattern center reference material, where the (114) pole serves as a decent approximation to the PC position.  

Practically you can run from 60-80 degrees, and as the tilt increases, your signal level generally increases, but you also exposure yourself to more uncertainty risks when calculating beam position due to tilt uncertainty at those values.

This is to increase the backscattering yield. 70 degrees also ensures a safe configuration in the SEM chamber, without posing dangers to the system.

To perform EBSD you need a double-scatter-event. First you need a quasi-elastic electron scattering to form a coherent and divergent electron source inside your sample. After that the electrons must rescatter at the atomic planes and come out of your sample. To ensure that you get a good SNR you need a large interaction volume and a very low distance between your virtual electron source and the sample surface. Both conditions need a high tilt angle. You can improve the SNR by using higher angles and you can improve the lateral resolution of EBSD by reducing the angle but that decreases your SNR so taht at about 55 deg the patterns vanish. Typically the EBSD system geometries are optimized for elements with avergae Z-values to ensure that the maximum of foreward scattered electrons hit the detector. In another geometry like nearly zero deg. you can theretically also perform EBSD-like techniques named SACP or ECP at single crystals. 

@ G. Qian: wouldn't it be better to vary the working distance to prevent any damage? Om the market we have many SEMs with different design related to chamnber size, pole piece geometry etc. When I interpret your answer correctly the 70 degree tilt is some kind of compromise between them?

@ F. Schäfer: it surprises me that you are talking about a large interaction volume when we take into account the launch of transmission Kikuchi diffraction. I thought the information volume (which is certainly not the interaction volume) can be even smaller for TKD (and so also for EBSD) than for diffraction in TEM?  

Do you have any idea why EBSD patterns vanish at 55 degree tilt?

How I have to adapt my EBSD system for elements (or phases) with non-average Z-values (which element numbers are concerned?) so that the maximum of forescattered electrons still hits the screen?

You can actually get EBSD-like patterns with a flat sample, but the contrast is reversed.  You have to lower the working distance and point your EBSD camera downward, but it's a fun experiment.  I've mapped a few materials in an older JEOL 6400, where we could easily lower sample and position camera.

The history, as I've heard, is that 70 degrees comes from using (100) Silicon as a pattern center reference material, where the (114) pole serves as a decent approximation to the PC position.  

Practically you can run from 60-80 degrees, and as the tilt increases, your signal level generally increases, but you also exposure yourself to more uncertainty risks when calculating beam position due to tilt uncertainty at those values.

@G.Nolze. I thought we talked about usual EBSD (70 degree tilt angle). The interaction volume I meant is the volume near the surface where electrons can still leave the surface not the whole volume where electrons can be found. I naturally wrote about the interaction volume for EBSD not for backscatter electrons. The idea is that for a grazing incidence angle the amount of back- or foreward scattered electrons is maximum. Naturally for high z angles I must increase the WD with a fixed EBSD-detector or if the detector can be moved one must move it to lower WD values. Low Z material like Al instead need low WD with fixed detectors. For t-EBSD the interaction volume of the elctrons is also the whole specimen thickness but only the electrons at the lower side of the sepecimen can get out of it. So even for tilt angles of about 20 degrees the SNR is improved without reducingthe lateral resolution significant. 

As far as I understand EBSD, only electrons of very similar energy ~+-250eV contribute to and EBSD patterns. All other electron of course also deliver a diffraction signal, but this is degraded in energy and therefore less visible. But it causes blurring of the diffraction signal. On the other hand it means that in reflection only electrons which mainly come from surface near areas (a few upt to a few ten nm) statistically fulfill this requirement since if they penetrate deeper the probability increases that you have additional scatter events so that this electron looses energy and cannot contribute to the pattern. The interaction volume is so not the information volume what is still the reason that BSE images not reach the same spatial resolution as EBSD patterns display. For TKD you have similar conditions. Only these electron which can pass the material nearly unscattered won't loose energy. On the other hand, if they are not scattered they have an excellent lateral resolution. And moreover, when they only display an interaction close to the lower surface they have the same or a similar z-resolution as EBSD, i.e. a few to a few10 nm. This is a big advantage with respect to TEM since there the entire material contributes to the diffraction patterns so that a FIB cut with a thickness of 80...120nm containing nanocrystals with a diameter of 10nm cannot investigated by TEM (at least not as single crystal) but for EBSD this is not a problem since only the last 10 or 20 nm deliver the extracted diffraction signal. 

Regarding the tilt of your sample: if you use a FIB section you can even tilt your sample nearly 90° if you like. Nevertheless, I see, what you want to say. In case your samples are too big they perhaps touch the pole piece. I myself perform nearly all measurements at a WD of 19mm. Of coures, for the rare situation of high resolution I shift to shorter WDs but then the samples are usually smaller or should be smaller anyway because of the stability of the sample position at high tilt. But if you are working with sample diameters of 10...15mm I don't see a problem to go to higher tilt (as Matt said).

I don't know exactly how you understand the term "grazing incident" but from my point of view, there you have no maximum of backscattered electrons, cf. Reimers SEM book 

I think we are all talking from the same things so I'm gonna leave this discussion. 

I think this paper gives some simple and easily understandable reasons for high tilt angles for the EBSD analysis (D.L. Barr, W.L. Brown, Rev. Sci. Instrum. 66 (1995) 3480-3482). Also, as EBSD signal comes from the top few tens of nm of the surface region, we need a 'grazing incidence (GI)' geometry to increase surface sensitivity - similar to many GI synchrotron radiation techniques?? I kind of agree with Florian in this regard, but I can be completely wrong. Furthermore, as also mentioned in the published paper given above, high tilt angle(s) ensures good contrast in EBSP - also check this out "http://www.ebsd.com/index.php/ebsd-analysis/ebsd-experimental-techniques/effect-of-sample-tilt."

Thanks for this interesting thread, I think I've gained something uesful out of the discussion! Cheers, GQ