Talk to the program director: pitch your research, ask for suggestions, ask how much money to budget, and read his or her body language (to see whether you should be applying to some different program or will have only a poor chance).
Read some (successful) CAREER proposals, especially in your own area. NSF makes the abstracts of all funded proposals available at http://www.nsf.gov/. This lets you see in which directorate and division your proposal belongs; also, you should ask some of your friends or colleagues to see their proposals. Get on an NSF review panel to see other people's proposals (CAREER or otherwise).
Don't promise to do more than you would otherwise (the panel won't believe you are superhuman), just that you will do new things, imaginatively.
Don't be shy about touting your own accomplishments. Use your record to support that you will be able to follow through on your promises, that it isn't just pie in the sky. You must build on what you have done before (to give credibility that you can do this), but also do something new and exciting, not just an obvious extension.
This is a planning document that extends for your entire career, not just four years. Have a vision, not just a list of individual actions. Show why people should believe you and believe in you.
Obtain supporting letters. In industry, it's good to get them both from research staff and from executives (such as a vice president). It's easier for them and less error-prone if you write the letter for them. It helps, but isn't required, for the letters to mention money (either past support or, even better, support contingent on you winning the CAREER award). These are not reference letters: they are about the research and only incidentally about the researcher. They should say how the writer will benefit from the research if it is supported.
In your application, mention any university matching grant (UW Engineering provides $5K if the department provides another $5K), and industry gifts (monetary or in-kind). Try to obtain such funding.
Have interdisciplinary links; if you don't already have these by the time you write the grant, it may be too late, but you should be thinking about it anyway. Being interdisciplinary at your core is dangerous, for that makes it easy to fall through the cracks; however, it is definitely a plus as an additional component.
Address diversity; for instance, at UW use the Women in Science & Engineering program.
Start early, especially obtaining letters and dealing with FastLane, which can be flaky. Get feedback from people who have written successful proposals in your area.
Hot topics are a double-edged sword; they get mentioned too much. If you do, have something substantive and new to say about them.
Ask for too much (i.e., the maximum) rather than too little funding (but talk to the program manager first); it doesn't hurt the proposal and you can always get whittled down but if you don't ask for enough, you'll never get more.
The education component is quite important; it should not be just what you would do anyway, such as teach classes and advise graduate students. Common ways to extend your reach include innovation in undergraduate education, outreach to high school students, continuing education, and writing a book. (Almost everyone includes a plan to develop new courses or involve undergraduates in research; you should probably say that too, but it won't be enough.)
Use the resources at the university to help with outreach (and show the potential impact of your research). Also say you'll put your courses on the Web so others can benefit (and do this with your old courses, for credibility); this won't help much, but it can't hurt.
Read about education, research in education, and the problems of the educational system; do your homework, have citations to educational research, and address the problems noted. See the NSF and ACM websites and the many panels on "What is wrong with education today". At UW use the Center for Engineering Learning and Teaching (CELT).
Make a powerful, concise presentation of your ideas and goals. Use figures; it clarifies the proposal, and some people will just read the figures. Don't use a tiny font; it overwhelms the reviewers. Use short paragraphs and short, punchy sentences. The project summary must be outstanding. Reviewers often read (only) the first few and last few pages of a proposal. Make it clear: use headings that tell not just what the topic is, but what the conclusions are. Tell a good story: both what and how. Don't neglect the evaluation component; give concrete evidence (including goals, milestones, and timetables) for what you are going to do. For a CAREER grant, the audience will be generalist, representing all of computer science, not just your field. A wise goal is to make one person on the panel fall in love with the work and the others not hate it; then your champion will win you the grant. A reader should be able to "feel the excitement rising from the page".
Back to Advice compiled by Michael Ernst.Michael Ernst