How to: write a scientific paper (in physics, anyway)

There are many resources out there for people who want guidance on scientific writing.  Here, in very brief form, are my tips.

• First, do some science.  That sounds flippant, but in order to justify adding to the scientific literature (assuming we're talking about a research-based paper here and not a review article), you do have to have done something.  Figuring out really what you've done and being able to articulate it clearly in a couple of sentences is essential.  What do you know now about physics that you did not know before the work?  Compelling papers are stories - they have some narrative structure (more on that below).
• Who is your audience?  Remember that you would tell your story differently to a specialized audience (e.g., your direct competitors who already know the topic in depth) than you would to a generic physicist in your subfield, and still differently to a generic physical scientist (if you're aiming for a broad/glossy journal).  This is equivalent to figuring out at least in rough terms where you are going to submit the work.  Knowing your audience gets you in the right frame of mind to....
• Figure out your figures.  Ok, so you've done some piece of physics.  How can you explain that in images/graphs/visualizations?  For a short "letters" paper, it's typical to have around four figures.  Often the first figure somehow sets the context and may include a cartoon/diagram of the setup or the system.  The figures need to tell the story in a logical way - if you're an experimentalist, show what you measured, and show how you got from those measurements to your conclusion (characterizing a new phenomenon?  Comparison with some candidate models?).    Again, who is going to be reading this - a specialist, or someone who needs some intro context?  Work hard on your figures - many readers will spend far more time looking at your figures that sifting through the text.
• Tell your story by describing the figures.  As I said, a paper is a narrative, but unlike historical literature you are not required to describe what you did chronologically, and unlike detective stories you do should not leave the main point for the very end.   If you've chosen your figures well, you are now something like 80% of the way there.  Do not worry about length at this point.
• Now work on the intro and the conclusions.  The intro needs to reflect the Big Picture and place your work in context, while citing appropriate literature references.  In a short paper, you don't need to cite everything remotely related to your work, but you don't want to leave out major contributors.  Don't cite yourself unless it's really germane.  The intro is where you will gain or lose the audience, including reviewers.   Usually at the end of the intro is a paragraph that starts out "In this work, we...."  That's where you need to be able to hit the highlights of your work in just a few sentences.  Really work on your intro.  It's right behind the figures and the abstract in terms of making an impression on the reader.   As for conclusions, what you do here depends on the journal.  Some journals just don't do concluding paragraphs.  Others wrap up the discussion, summarize the results again briefly, and then give perspectives on possible future work.
• Lastly, do the abstract (though I often rough one out earlier).  Again, the abstract needs to be concise, clear, provide context, and summarize the main points of the paper.  As annoying as it is on some level that there is an algorithm for this, there really is, and it works.  Take a look at the Nature guide for writing an abstract (pdf!).   The general format and the way it makes you think about what each sentence is doing can be very helpful, even when generating abstracts for non-Nature journals.
• Now edit.  Read and re-read the paper, rewording things to make them more concise and clear.  Put it aside for a day or two and then look at it with fresh eyes.  Make sure the coauthors do edits as well.  My approach is definitely to rough some text onto the page and then edit, rather than agonize about trying to write something perfect from the outset.  
• Think about a cover letter for the submission, and come up with some suggested referees.  For the cover letter, you are trying to convey clearly and cleanly what you did to an editor who has to filter through many many papers.  Don't make it a lengthy exercise.  Do make it accessible, and do make it clear why your work is appropriate for the journal.  On the referee front, don't try to stack the list with your pals.  Pick people who will actually give insights and have relevant expertise.  The editor may well take your suggestions as a jumping off point for picking people out of their database, so if you pick people who have a clue, it increases the likelihood that your referees will be thematically appropriate.

Good luck!  Feel free to add in more suggestions in the comments; I'm sure I'm leaving things out.