I have collected here some guide lines on how to give technical talks. The first one is from a periodic comp.usenix posting by Henry Spencer. I've found that his points are applicable to a wide variety of situations.

The second set of guidelines is by Jim Blinn, and is from the SIGRAPH 2000 web pages. It was originally at http://www.siggraph.org/s2000/presenters/papers/jim_blinn.html. I've reproduced it here in full, just in case the SIGRAPH 2000 pages go away. As you might expect, The Blinn guidelines pay more attention to graphical details.

Guidelines for giving talks by Henry Spencer:

1. Usenix is coming up fast. In a continuing effort to make conference papers more worth listening to :-), some thoughts for the presenter, especially the first-time presenter... (These are my own opinions as a long-time attendee and sometime presenter, and don't necessarily reflect the views of Usenix or anybody else.)
2. Run through your talk at least once in advance so you know how long it will take. It probably needs to be cut down. My talks usually need to be cut by at least a factor of two. Doing this in advance is far superior to improvising when you discover you only have five more minutes. Don't forget to leave time for questions at the end.
3. Doing a practice run in front of a mirror can be revealing. Mumbling, speaking to your notes, etc. can be fought effectively by speaking only when you have eye contact with a member of the audience. Not the same person all the time, though!
4. Avoid wah-wah syndrome: if you're wearing a lapel microphone, don't keep turning your head to look at the screen while you're talking, because it makes your voice fade in and out. Put your notes on paper, not on your slides (or make a paper copy of your slides), so you can have your notes in front of you while you talk.
5. Say something interesting. If you're short of time, pick one subtopic and say something interesting about it, and forget the rest. Few things annoy an audience more than hearing "well, the rest of that's in the paper" every two minutes. Especially since it usually comes just as you were starting to talk about something in enough detail to actually be worth listening to. If there isn't time to cover everything well, don't try. (This is especially true if you find yourself unexpectedly short of time. Dashing through half a dozen topics in five minutes means you can't say anything interesting about any of them.)
6. Say something your competitors would find fascinating. If they wouldn't find your talk interesting, the rest of the audience probably won't either.
7. Remember that your audience can read. Rather than reciting the results in the paper, talk about how you got them, especially what problems you solved and what mistakes you made, or about the implications.
8. Plan your slides for visual content. Pictures should be worth a thousand words. If a slide isn't useful when shown backwards, it has too many words and not enough picture -- take it out. The outline of your talk belongs in your notes, not on your slides.
9. Using visual media for things with visual content also does wonders for the poor people at the back of the audience who can't read your 3-point type anyway. If you must put words on your slides, 24-point type is about the minimum. For viewgraphs, stand up and drop them on the floor at your feet; if you can't read them that way, the print's too small.
10. Visit the washroom shortly before your talk.

Things I Hope Not to See or Hear at SIGGRAPH

The following is an excerpt from the indispensable book: "Jim Blinn's Corner: A Trip Down The Graphics Pipeline"

No, I'm not going to talk about flying logos or glass balls. I am going to talk about that special form of performance art known as "Giving a Technical Presentation." These ideas apply to speakers in panels and tutorials as well. I realize that there is a somewhat small direct audience for this, but others of you might be able to use this information in your own talks elsewhere. Also, you should expect this from presentations you hear at SIGGRAPH.

SIGGRAPH sends out a lot of stuff about how to prepare visuals, etc., although, from what I see, not many people read it. Reading this chapter does not excuse you from reading SIGGRAPH's materials though. The following ideas are just my own personal biases. I will phrase them as things not to say/do because, let's face it, it's a lot easier to complain.

Talks Read Verbatim
A technical talk is just one facet of a multimedia event built on your work. An adventure story appears different in the film version and the book version. Likewise, different things are appropriate for the spoken version of your paper than for the printed version. A much more conversational style is best for the talk. Tell a story about what got you interested in the problem in the first place. Briefly relate some dead ends that you tried that didn't work. But please don't read your paper verbatim. We are people out here in the audience; we're all your friends, just talk to us. The only exception to this rule is if you are not a native English speaker. If you are not fluent in English, it is probably best to have your words already prepared.

Illegible Slides
The most important part of your talk is the visuals; this is SIGGRAPH after all. I am sometimes amazed at how many illegible slides are shown, most especially by representatives of organizations (who shall remain nameless) that sermonize about high-quality imaging. Here are some things that have disturbed me most about slides I have seen.

Microtext
Many of you are involved in the microcircuit revolution and tend to think this also applies to the text on your slides. It doesn't. My personal rule is to put no more than six lines of text on any one slide. And while you're at it, use the biggest font you can that will fit on the slide. Six lines of teeny-weeny text with gigantic borders is still not readable. But, you may ask, what if I have more than six lines? Well...just use more than one slide. See? Simple.

A good check for readability of slides is to hold them at arm's length and see if they are still readable. (That is what I do, and my arms are probably longer than yours.) Believe me, that is how small they look from the back of the room. In fact, I make all my slides on my animation system that only has video resolution. This may seem to be a disadvantage, but it's not. It forces me keep the slides simple enough to be legible from a long distance.

One effect of this restriction concerns equations. You simply can't have a complex equation on a slide. Even if you shrink its many terms down so they will fit, it will look like grey noise from the back of the room. Recast your equations into simpler chunks and give each chunk its own name. Make one master slide with the basic equation in terms of these names. Then make a separate slide to define each chunk. Don't put more than one equation on a slide unless it is fantastically necessary. Use separate slides for each equation; it focuses attention while you are talking and gives you more room for one.

Magenta Lines on a Cyan Background
Another design issue concerns colors and contrast. Your best bet is to use some dark background (like blue) with very light color text (like white or yellow) on it. Alternatively you could use a light background and dark lines. Even then, I have seen some terrible slides that use black letters on a white background. Even though the letters were big, the slides were illegible because the lines were too thin. Light areas seem to expand visually, so dark lines tend to get eaten up by a white background. If you must use light backgrounds, use a much thicker line width for the dark lines to compensate for this phenomenon. If you want to emphasize some items on the slide, make them in a lighter color than the rest (not just in a different color).

The Entire Text of the Talk Echoed on Slides
The audience is not going to want to read a lot of text while simultaneously trying to pay attention to what you are saying. Text on slides should just consist of section headings. If you have a section of your talk that you don't have any obvious graphics for, don't feel compelled to put the text you are reading on a slide just to have something there. The days of silent movies are over. If you must have something, try showing a picture of a pretty waterfall.

And remember, folks, no overhead transparencies allowed. There is a reason for this: they look terrible no matter what you do.

"I'm Sorry These Slides Are So Dark."
I don't think I have ever seen a slide at SIGGRAPH that is overexposed. When you film your efforts, make several exposures and pick the brightest one. In general, err on the side of overexposure; make the exposures longer than you think will be necessary.

But for heaven's sake if, despite my sage advice, your slides don't come out bright enough, don't make a big production out of apologizing for them. It doesn't make them any more readable, and it may just call attention to problems that may not be as noticeable as you thought. Your view of the slides from where you speak is not the best one. The slides will look a lot brighter to the audience than they do to you. Just show them and get on with the talk.

Likewise, don't spend a lot of time fiddling with the focus (which requires shouting at the AV people in the back of the room). In the first place, your slides should be big and bold enough that a little bit of out-of-focus shouldn't bother them. Remember, from the back of the room the screen looks like a postage stamp. Problems with focus that appear bad to you, with your nose three feet from the screen, won't show up to the audience.

Talking about, and taking time with, these issues distracts from your presentation.

The Floating Head
Because of the size of the auditorium, your face will probably be televised on a large TV screen behind you. The AV people set up the lighting to make your face optimally visible. This often has the effect that dark-colored clothing completely disappears into the background. This gives the impression of just your face floating in a sea of black. So...wear light-colored clothing. Your shoulders and arms will then show up and your audience will be able to tell that you are a whole person.

The Tops of Speakers' Heads
No, I'm not saying this because I'm tall. I mean that speakers should look straight out at the audience instead of burying their noses in their notes. I know it looks like a black hole out there, what with the dim house lights and the spotlight on you. You can't really see the audience, but there are people there. If you look down all the time, all that people will see is the top of your head. This is so important that I'll say it again. Look up at the audience; it looks a lot better for the TV cameras.

Also, don't turn around to admire your face on the big TV screen. It just won't work; all people will see is the back of your head. Likewise, don't turn around and look at your slides all the time (except maybe for a brief glance to make sure you are on the one you expect). Traditionally, people really are more used to seeing the front of peoples heads than any other side.

The Fading Voice
Another reason not to turn around a lot is sound. There is a microphone in front of you, not behind you. A lot of speakers start out saying something to the microphone like "And as you see in this slide..." Then they turn around and look at the slide and say "...the secret of the universe is revealed." Only they aren't speaking into the microphone anymore. What comes out is "mumble mumble mumble." Speak consistently into the microphone. Let the your secrets be revealed.

Wiggly Pointerism
You will probably have a laser pointer to use during your talk. Since the screens are so large and so far away from you, a very slight motion of your hand will make the pointer jump around in a very distracting fashion. Try to keep your pointing hand as steady as possible to keep the audience from getting seasick. Or else turn it off when you aren't actually pointing at something.

"I'm Almost Out of Time so I'll Just Run Through the Rest of These Slides Real Fast."
You are hereby warned: you only have about 15 minutes to do your brain dump (for a talk in the technical session). The time you have is well known to you in advance; you must use it wisely. About all you can expect to do in this amount of time is give an overview of your paper and inspire those in the audience to read the paper itself for details.

Plan on spending most of your time talking about your new ideas. I have seen talks where the speaker spends 13 minutes giving a review of the field and a justification for why their specific problem is interesting. Then---what do you know---there's no time left for the meat of the talk. I think you can safely assume that most everyone in the audience thinks computer graphics is a good idea and that, in fact, the specific problem you are addressing is worth solving. You can probably do fine with about two minutes of introduction before getting to the good stuff.

Don't go into enormous detail in derivations of the math, just give the basic assumptions and the results. This simplification process goes hand-in-hand with the simplification of your equation slides. The general gist of the math should be describable without going into a lot of fine details that people will best get out of the paper.

If you have a videotape, time it and make sure it doesn't eat up the whole time for the talk. Speaking from experience, it is very embarrassing for a session chairman (whose main duty is as time police) to have to interrupt a nifty tape because there's no time left.

"Uh, I Guess That's All I Have to Say."
Probably the most important parts of your talk are the first and last sentences. Have these all figured out before you go up to the podium. Try to have something snappy to end with rather than just drizzling off. You also must give the audience a signal for when to applaud. Usually a simple "Thank you" will suffice.

Remember
Look up. Bright slides, big letters.

Uh, I guess that's all I have to say.

Thank you.