This is a vast subject, and probably has been covered by many articles (or even books). But here let me focus on the most crucial and fundamental issue.
Theorem
The nature of a research presentation is to convey ideas that you know to others who do not yet know.
Corollary: no need for stage fear
There was a study indicating that people fear public speaking more than death. Not everyone has this issue, but if you happen to be nervous about an upcoming presentation, just remember the basic theorem: you know much more than your audience about what you plan to talk about. So, even if you make mistakes, the audience would probably not find out anyway. Just stay cool, and do not let your tone, facial expression or body language disclose the fact that you are screwing up.
Corollary: never over-estimate your audience
Unless you could read mind, it is probably very difficult to devine what other people do not know. Since you already know what you want to talk about, it is very tempting for you to recite what is already in your mind instead of what your audiences really need. This is the single most important cause for a bad presentation in my personal observation. There is no magic solution, but a useful heuristic is to never over-estimate your audience. Always start your presentation with the basics; if the audiences do not know that, they will appreciate your effort, and if they do, they will feel they are smart. Convey high level information instead of details, and use easily absorbable medium, like images or videos, instead of texts, to convey your points.
Corollary: never over sell
The goal of the presentation is more to entice people to be interested in your research than to teach them the details. In a sense, it is more like an advertisement than a class room teaching. It is nearly impossible for average humans to learn a new subject within a 20 min presentation. So do not try to cram in every single detail of your algorithm into your talk; probably nobody is going to get that anyway. Instead, focus on getting the audiences’ attention for the first 10 minutes of your talk. And if you could achieve that, you are on your way for a great presentation.
Finally, like many other aspects of research, the best way to learn is by experience. The more presentations you give, the more likely you will learn how to give a great one. Do not worry about failures; I totally blew up my first research presentation, but eventually I figured out the deals, and now I am not only highly comfortable but also highly enjoy giving research talk, especially to huge audiences like SIGGRAPH.