Most of this advice applies equally to proposal writing, but the purpose differs. You are attempting to gain endorsement from, as well as create understanding in, the reader. Hence it must be tailored to your target audience (the reviewers) and you must motivate them. Your target audience will depend not on your choice but on your choice of the sponsor and the sponsor's mechanism for review. There may be more than one audience (peers plus program manager) or only one person (program manager). Find out what the target audience will be before you write the first word.
Questions that no reviewer should have to answer are only a few, but they are very important. What are the current, exciting problems in your field (does the proposal writer have a clear grasp)? Why does the one that you have chosen deserve to be on top or near enough for funding? Why should this work be done now, i.e.: why has it not been done yet, what bad things will happen if it is not done next and what good things will happen as soon as it is done? Why can this work be done now (convincing assurance of feasibility)? Why or how are you well poised to address this problem?
Feasibility is perhaps the biggest issue, even if you succeed in convincing reviewers of the importance of the problem. For peer review, cite your past, relevant successes and show preliminary results. For review by one or more program managers a track record of success with other tough problems may or may not substitute.
When writing to a program manger alone, keep it short. When writing to peers, follow the rules on lenght in excrutiating detail, including font faces and sizes. You don't need extra complaints. Because they are read early and often, concentrate on the abstract and introduction.
Avoid using a potential reviewer as a strawperson. If you have something nice to say about an idea, feel free to associate it with an author (unless he or she has many enemies or unless you might err in attribution). If you must criticize an idea, separate the idea from the author (give it a descriptive label rather than "Jones' hypothesis"). Give only enough references from your lab and your close associates' labs to demonstrate that you are competent; cite potential reviewers' papers whenever possible in place of yours. Demonstrate command of the literature by citing key reviews, key papers and other highly relevant work by likely reviewers. It is far dumber to misquote a paper written by a reviewer than to omit it.
Get the big idea out clearly, early, and often, with minimal jargon. For peer review in particular, demonstrate that you can frame and manipulate technical and testable hypotheses. The more precise the structure of your hypotheses, the more they will satisfy the other end of the spectrum of reviewers from those who want the big picture, so you will have satisfied both groups.
Deal with real and perceived overlap explicitly. Make clear how your proposed work differs from that in other labs. Make clear how your proposed work differs from other current and proposed work in your lab.
Help the reviewer do a reality check. Make clear the minimal deliverables. If you fail to deliver and have to propose again for the same or very similar work, people will assume that you have failed. One-year grants are particularly treacherous because you will be writing proposals again before you have much to show. So, don't ask for too little money or too little time (duration of grant). Don't pad with requests for unrelated resources. Make clear those items without which you can't live. For a preproposal or letter of intent, guess high, not low: you might get the OK. Program managers love to hear about price decreases. Program managers hate creeping inflation and the people who cause it. In a budget, list personnel by name whenever possible; it is much easier to cut an anonymous position than a real person.
Show respect or suffer the consequences. Follow the agency’s format precisely. Bulletize your main points, preferably in words that can easily make their way into the reviews. If review criteria are explicit, use them as a checklist. Check each proposal copy yourself; the reviewer will not blame your secretary, administrator or Kinko's. If submission is electronic, send it off and then print it from the source that a reviewer would use; proofing your original is not sufficient to avoid formatting problems. Include a few attractive and relevant illustrations. (So few people do that it is unexpectedly effective.) Avoid first-person plural unless you have a Co-PI. "We" and "our" invite disagreement, especially in proposals.
Don't take rejection personally or hard. It pays, literally, to be thick skinned and persistent. Review is a stochastic process based on a small number of returns. The precision of reviews is simply awful (i.e., there often is no statistically significant difference between a mean rating of excellent and good). If you don't play, you can't win.
Do take reviews and reviewers’ time seriously. Reviews tell you where to spend your time on improvement the most profitably. Don't resubmit to the same agency without responding to the prior reviews because reviewers will (and should) scream, "Bloody murder." A review rating the work as excellent can kill you, if it points out a serious problem. Don't pay exclusive attention to the reviews that give low rankings. Ask the program manager point blank for her or his statement of the most serious shortcomings. Persist until you make direct voice contact. No matter how uncomfortable for you, it beats the alternatives. There are many important things that people wont and shouldn't put in e-mails. Alternate two proposal topics to an agency, if necessary, to avoid resubmitting a proposal that still contains serious problems already pointed out in previous reviews.
The advice so far I doubt is very controversial, but my advice on interdisciplinary proposals may be more so. By definition, interdisciplinary proposals fall between two established disciplines that have numerous competent reviewers (as well as the other kind). Important questions differ among disciplines. There are very few genuinely interdisciplinary questions. (I can't think of any.) If you are lucky, you will find a suite of interactions where the discipline playing the service role alternates (e.g., sediment transport effects on organisms vs. organism effects on sediment transport), to keep the problem interesting to both sides. There are many good questions within disciplines that require help from other disciplines in getting an answer. For example, to find out whether year classes are missing due to lack of larval return to a site requires physical oceanographic information. Face it: The needed physical information to solve the biological problem is not likely to be rated highly in terms of providing new physical insights; the space and time scales are set by the organisms development and therefore are not likely to match the physicist's preferences.
Don't expect high ratings from the service discipline. Before you write a proposal or manuscript, identify the reviewer base who will endorse your question as important for their field and write to them, not to some imaginary guru between disciplines who knows it all. If you do not establish this sort of reviewer base within one discipline, you will not be free to engage in interdisciplinary work because you will not get proposal funding, manuscript acceptances or gainful employment as a proposal-writing faculty member. If you can't identify at least 5-7 people within a discipline who would class the work as important, you need to change your question or your strategy. Only once you have established a reviewer base within a discipline or a willing program manager in a non-peer-review system are you free to take greater risks in the territories between disciplines.
This advice holds within biological oceanography as well as across its boundaries. For example, take a scientist (e.g., John Hobbie, Fred Dobbs or Craig Plante) who first made an impression as an invertebrate zoologist before deciding to go in the direction of microbiology. In a peer-review system (NSF proposals, manuscripts) it is a mistake to think that you will get the benefit of the doubt from reviewers who view their expertise as being different from yours. In a peer-review system, get funded or published on a solid reviewer base, and then experiment cautiously with branching out. If you want to switch fields, associate yourself initially with an authority in the target field — one who has a solid reviewer base. Get your union card. Especially during times of poor funding, peer reviewers are not likely to give the benefit of any doubts.
Other approaches than writing for a specific reviewer base do sometimes work, but they are like playing the lottery with even worse odds. Send your first proposal as a PI to a carefully targeted reviewer base.