ACM's highest membership grade is Fellow, intended to recognize and honor outstanding ACM members for their achievements in computer science and information technology and for their significant contributions to the mission of ACM. The goal is for Fellows to represent the top 1% of ACM's membership. (The other advanced membership grades are Distinguished and Senior.)
I served a five-year term on the ACM Fellows Committee, including chairing it one year. And as co-chair of the ACM Awards Committee, I have been a non-voting observer for the past seven years. In those twelve years, I have observed a number of common problems with candidates' cases that I can discuss without violating the confidentiality of the committee's discussions. 
First, some preliminary remarks. This is a very dedicated and hardworking committee, who take their task very seriously. They are unpaid volunteers who each year read and consider some 700 nominations and endorsements. Although their collective expertise covers an impressive portion of the discipline, it is inevitable that no member of the committee will be at home in the areas of all of the candidates. But they each try to properly evaluate all of the cases presented, prior to a face-to-face meeting where they discuss each case.
Fellow candidates are not in competition with one another. There is no annual quota; each case is judged against the qualities of previous Fellows and the committee's subjective opinions of what constitutes "the top 1%." Although the roster of Fellows is approaching 1% of ACM's professional membership, the full 1% is unlikely to be reached any time soon, due to a combination of ACM's professional membership growth and attrition among the Fellows.
Fellow cases fall into three groups of roughly equal size:
Cases generally agreed to demonstrate that the candidate meets the criteria.
Cases generally agreed to not do so.
Cases where there is either disagreement or uncertainty within the committee.
Not surprisingly, the bulk of the committee's discussion is devoted to the third group, and this is where the chance of mistakes is greatest.
Every year there are surely a few candidates who are rejected, not because they are not qualified, but because their nomination and endorsements fail to make the case as effectively as possible. There is no way for the committee to know which cases these are. When its discussion is inconclusive, the committee tends to make the conservative choice: A candidate who is not accepted, but deserves to be, can be re-nominated; a candidate who is accepted will not be re-evaluated later. 
Here are my suggestions for nominators and endorsers about what they can do to strengthen (or avoid unnecessarily weakening) the cases for their candidates.
Nominate the candidate: Probably the most common reason for people who should be ACM Fellows not being accepted is that they have not been nominated. The committee is not empowered to nominate candidates on its own, it can only judge the nominations it receives. 
Follow the rules scrupulously: ACM's software and its Headquarters staff check all candidate packages strictly. If a nomination or endorsement arrives after the deadline, the committee doesn't see it. If a nominator or endorser isn't an ACM member as of the deadline, the committee doesn't see their input. If there are not five endorsements, the committee doesn't see the case at all. Etc.
Summarize the case in the nomination: In addition to supplying the required data, use your limited space to describe the candidate's accomplishments briefly, but specifically.
Avoid flowery adjectives, multiple superlatives, and overworked words such a "seminal" and "unique." Instead, indicate the importance of the work by describing its specific consequences (e.g., a new line of research taken up by others, a system or product in widespread use, an important conjecture resolved, a popular textbook) in terms understandable to a non-specialist.
The ideal candidate will have distinguished personal technical accomplishments, distinguished technical leadership, and distinguished ACM and community service. The more "concentrated" the accomplishments are into one or two areas, the more outstanding they need to be in those areas.
Although there is no formal age or experience requirement for Fellow, evidence is needed that the candidate's accomplishments already--not just potentially--rank in the top 1% of ACM's membership.
Cases with a weak nomination letter usually end in the bottom third and get little discussion.
The criteria for ACM Fellow are different from those for academic tenure: A recycled tenure case, no matter how strong, is not likely to be effective.
Choose the endorsers carefully: Since it is assumed that likely candidates will all have strong nominations, much of the committee's discussion focuses on the endorsements.
Each endorser should be able to personally attest to portions of the candidate's accomplishments.
An endorsement from an ACM Fellow carries just a little more weight, provided that the Fellow is actually familiar with the accomplishments discussed.
An endorsement by someone known to members of the committee can be more readily calibrated.
It is helpful to have endorsers from several different organizations, and endorsers who can address in more detail each of the areas of accomplishment outlined in the nomination.
With few exceptions, those who endorse more than two or three candidates will not strengthen their cases. Nor will cycles in the endorsement graph.
It is acceptable to seek more than the minimum of five endorsements, to guard against the contingency of an endorsement arriving late. But the committee will see just the first five endorsements that are submitted, so try to make sure those are the strongest you can get.
Write the endorsements as carefully as the nomination: An endorsement should focus on accomplishments that you can personally attest to, place them in context, and explain their importance. If you can, include more specifics than are in the nomination. Merely repeating portions of the nomination letter will actually weaken the case.
If you are endorsing multiple candidates, it is important to rank them, and useful to compare them. (If you cannot do this, you probably don't know all of them well enough to endorse.)
Notes:  This post is an update of an earlier post on my personal blog.  Nominators of rejected candidates should seriously consider whether a stronger case could have been made; if they are convinced it could have been, it is acceptable to re-nominate the next year (but not effective to do so with the same letters). If a candidate is rejected two years in a row, a third consecutive nomination is not allowed. Wait a couple of years; if the candidate then has further accomplishments to strengthen the case, consider re-nominating.  And, as a matter of policy, committee members recuse themselves from nominating or endorsing candidates, as do the Awards Committee co-chairs.