How to Get a Grant (from NSF, for Computer Scientists)

Learning to write good grant proposals is hard. First, while we learn to write papers in grad school, most of us only learn to write grants after completing a PhD, and possibly only after getting a faculty position. Second, the hit rates for grants are typically lower than for papers. Third, it’s fundamentally harder to write compelling text about a problem that you have not yet solved. For similar reasons, the PhD proposal is a major stumbling block for many folks. Fourth, grant review panels are smaller and have much higher variance in expertise than do top-tier conference program committees. Finally, there is less visibility into the proposal evaluation process than there is into the the paper evaluation process.

The upshot of all of these difficulties is that writing grant proposals is a bit of a frightening activity. One can spend a lot of time writing proposals before getting a hit. This is frustrating and, if it goes on very long, career-limiting. (Tenure committees usually evaluate candidates on research, teaching, and service — but all too often money is the ten-ton elephant hiding in the corner.)

This writeup is random thoughts, it’s in no way complete. You have to read the call for proposals and parts of the GPG carefully. You should talk to the program managers. You need to figure out how to work with your sponsored projects office and with your department’s internal grant-writing support people. Others have put useful information on the net; I found Cindy Grimm’s advice to be particularly good.

Writing a Proposal

To be funded, a proposal has to receive good reviews from respected researchers in your field. To get good reviews, there are three main things you need to show: a vision, a plan, and a capability. Let’s look at these in more detail, but first note that these don’t really match up with the parts of an NSF proposal as described in the Grant Proposal Guide. Rather, these are things that need to come out of the proposal as a whole.


The median annual income for a household in the United States is around $50,000. When you write a grant proposal, you’re typically asking the government to give you 2 to 100 times this amount. Think about it: every dollar you request will have to come from some hapless taxpayer (or corporation, or increase the debt…) and additionally won’t be used to cure cancer or put a person on Mars (unless, of course, your work directly attacks one of these problems). My opinion is that when you ask for this level of funding, the request had better be backed up by a damn compelling vision for how the world will be different if the money is awarded.

The vision part of a proposal is tricky. It has to be concise — probably around a page — and compelling, while remaining strongly connected to the proposed work. You can’t spin a story about ending disease and poverty, then immediately segue into open problems of the 5-bit frobnicator. The best vision parts of proposals leave panelists feeling that they have a mandate: the problem is so important that the work has to be funded no matter what. I’ve read proposals like this, they’re rare and wonderful. They always get funded.

In summary, the purpose of this part of a proposal is to convince your review panel that the problem that you are attacking matters, that it needs to be solved. It must not make people bored or irritated.


Here’s the really hard part: you have to convince the proposal review panel that you have an approach: a plan of attack for your chosen research problem. You must share your ideas, insights, and early results in order to paint a picture of a plausible research program that will come to fruition in 3-5 years. This is not the place to hold back on your most ingenious idea for fear of having it stolen. In general, it’s hard to find a piece of work that is plausibly doable, but that doesn’t sound like more of the same. Panels, in general, are not very interested in funding more of the same, they want new approaches.

One interesting question is: How much work should you do before writing a proposal? I’ve heard people say that you should basically have finished a piece of work before writing the proposal about it, and then you use that grant to fund the next piece of work. This seems pretty dishonest and in any case this approach would be highly risky if you’ve published the work. On the other hand, the reality is that it’s pretty tough to write a convincing proposal without having some kind of preliminary results. So I guess in the end the answer is pretty simple: you strike a balance that feels right.


A good proposal makes it clear that you are the best person in the world to do the work that you are proposing. You absolutely do not want a panelist thinking “Gee… that’s a neat idea, but who’s this schmuck? I bet Jones at MIT could knock off that project in six months.”

The capability to do research can be demonstrated in different ways. First, even if you’ve never received NSF support, you should briefly describe your previous accomplishments in the area of interest. Second, you might want to find an excellent (and probably more senior) collaborator who can be a co-PI on the proposal. Third, you can demonstrate capability by association, whether it is with equipment such as the huge cluster of Cray-27 nodes available only at your institution, or with people, such as your unique industrial contacts.

It can be daunting to realize that your competition may have been working in the field for 25 years, may run an operation with a budget 20 times larger than yours, and may have won every award the there is to get. In fact, these people often write incredibly polished proposals. However, it is highly likely that established people are working on established ideas. It is only the very best people in the field who keep doing extremely original work year after year, and you might as well just hope you don’t come up against them too often. In contrast, as a new person in the field you should have new ideas and approaches to bring to the table. Play this up in the proposal: it’s not just business as usual.

Other Issues

It’s important to make life easy for the people who will be evaluating your proposal. Assume yours is buried in the middle of a pile of about 20 proposals, and it will be read (probably on a plane to DC) by an exhausted person who would rather be doing something different. To make things easy for reviewers, keep it simple. Don’t assume everyone is interested in the deepest mathematics that you are capable of writing about. Use plenty of figures and graphs to break up the flow of text. Since some panelists like to skip around, try to make parts of the proposal as self-contained as possible and use informative section titles. Don’t refer to the previous work as nascent or vacuous since probably that work’s authors are on the panel.

Many of your colleagues have served on NSF panels. Get them to read your proposal and give feedback. Unfortunately, this will require that the proposal not be coming together on the day of the deadline.

Getting a Grant

Ok, so far this post has actually been about how to write a proposal, now let’s talk about how to get a grant. Realistically, unless you are both very good and very lucky, you need to submit more proposals than you are really interested in writing. One of my colleagues (not jokingly, I suspect) talks about submitting them faster than they can be rejected. Writing lots of proposals is sort of a bummer since each one keeps you from doing something else important, like writing a paper. Also, you never, ever want to write a proposal for work that you don’t actually want to do, since of course that proposal would end up being the only one funded.

When a proposal is rejected there are several things to keep in mind. First, it may not have been reviewed by any experts in your exact field. Second, if you revise and resubmit the proposal, it will be evaluated by an entirely new panel. A friend of mine once had a proposal rejected, but with strong reviews — it was barely in the not-funded category. He diligently revised the proposal and the next (much better) version got only lukewarm reviews. Again, he revised and resubmitted and the third time around got quite negative reviews. This kind of experience is not terribly uncommon and for this reason a different friend of mine makes a point of never reading his NSF reviews at all (or so he claims).