Reverting is reversing a prior edit, in whole or in part. Revert vandalism upon sight but revert an edit made in good faith only after careful consideration. It is usually preferable to make an edit that retains at least some elements of a prior edit than to revert the prior edit. Furthermore, your bias should be toward keeping the entire edit.
Reverting tends to be hostile, making editing Wikipedia unpleasant. Sometimes this provokes a reciprocal hostility of re-reversion. Sometimes it also leads to editors departing Wikipedia, temporarily or otherwise, especially the less bellicose. This outcome is clearly detrimental to the development of Wikipedia. Thus, fair and considered thought should be applied to all reversions given all the above.
The main purpose of reversion is to undo vandalism or other disruptive edits. If you see an edit that you're sure was intended by its author to damage Wikipedia, and it does, there is no need for further consideration. Just revert it.
In the case of a good faith edit, a reversion is appropriate when the reverter believes that the edit makes the article clearly worse and there is no element of the edit that is an improvement. This is often true of small edits.
Whenever you believe that the author of an edit was simply misinformed, made a mistake, or did not think an edit through, go ahead and revert. If that editor (or anyone else) re-reverts, you will know it is more than that, and you should be more conservative in deciding whether to revert it again.
Another kind of acceptable reversion is an incidental one. A Wikipedia editor is not expected to investigate the history of an article to find out if an edit being considered is a reversion of some prior edit. The rule against reversions applies only to cases where the reverter is aware that the edit is a reversion of another edit.
There are a number of things that sometimes motivate an editor to revert, but should not.
The first and foremost alternative to reverting when you find you disagree with an edit is to find a third version of the text that incorporates at least some of the elements of the prior text and the current text. Sometimes that's as easy as making the article state that there is controversy about something.
You might discuss an edit on the talk page before reverting. But note that Wikipedia does not in general require advance approval of edits, and reversions are no exception. If you believe you have a case of an acceptable reversion, you are invited to make that edit unilaterally and if there is disagreement, you'll find out from subsequent edits. (But note the special rules for avoiding edit wars.)
You could also discuss an edit directly with the editor who made it, on that editor's talk page, and request that the editor modify their own work. Or convince you that it's best as it stands.
High-frequency reversion wars make the page history less useful, make it hard for other people to contribute, and flood recent changes and watchlists. Wikipedia policy forbids anyone from reverting any single article more than three times within a period of 24 hours, with certain exceptions. This is a strict limit, not a given right; you should not revert any one article more than three times daily. Violation of this rule may lead to protection of the page on the version preferred by the non-violating party, blocking, or investigation by the Arbitration Committee. Usage of sock puppets attempting to circumvent this rule does not prevent a violation. See Wikipedia:Three revert rule for details on this.
Being reverted can feel a bit like a slap in the face – "I worked hard on those edits, and someone just rolled it all back". However, sometimes a revert is the best response to a bad edit, so we can't just stop reverting. What's important is to let people know why you reverted. This helps the reverted person because they can remake their edit while fixing whatever problem it is that you've identified. Obviously it is best to fix the problem and not revert at all.
Explaining reverts also helps other people. For example, it lets people know whether they need to even view the reverted version (in the case of, e.g., "rv page blanking"). Because of the lack of paralanguage online, if you don't explain things clearly people will probably assume all kinds of nasty things, and that's how edit wars get started.
If your reasons for reverting are too complex to explain in the edit summary, drop a note on the Talk page. A nice thing to do is to drop the note on the Talk page first, and then revert (referencing the talk page in your edit summary), rather than the other way round. Sometimes the other person will agree with you and revert for you before you have a chance. Conversely, if someone reverts your change without apparent explanation, you may wish to wait a few minutes to see if they explain their actions on the article's talk page or your user talk page, or contact the editor and ask for the reason for their revert. Do not engage in discussions in edit summaries. Doing so is a hallmark of edit warring; instead, stop editing and use the talk page.
Having realized that article development has ground to a halt because of incessant reversions, two or more people may agree to give higher-than-usual respect to each other's edits. Unlike the three-revert rule, these rules are usually voluntary and self-enforced. For cases where they are imposed as restrictions, see Other revert rules in the edit warring policy.
Some editors may choose to voluntarily follow a one-revert rule: If you revert a change and someone re-reverts it, discuss it with the re-reverter rather than reverting it a second time.
In a situation where two editors disagree, if both follow this rule, the history will show two reversions (one by each editor) before discussion starts.
Sometimes, users may be limited to one revert by the Arbitration Committee or by the community per the editing restrictions guidance. In other cases, articles or entire topics may be placed under a standing 1RR restriction.
Editors with this pledge choose to voluntarily follow the rule that if someone reverts any change of theirs, they don't re-revert it, but discuss it with them. (See Proposal.)
In a situation where two editors disagree, if both follow this rule, the history will show one reversion before discussion starts.
Editors may also choose to adhere to a zero-revert rule, for example:
"Only revert obvious vandalism. Instead of removing or reverting changes or additions you may not like, add to and enhance them while following the principle of preserving information and viewpoints. If you can't figure out how any part of an edit benefits an article ask for clarification on the article's or the editor's discussion page."
In a situation where two editors disagree, if both follow this rule, the history will show no reversions before discussion starts.
Using a zero-revert rule gives fellow editors the benefit of the doubt in all cases. Even in instances where you know the other editor's viewpoint is dead wrong, the fact that some people have this viewpoint can be relevant in itself, and their contributions might be expandable into a useful addition to the article. This rule can be very difficult for some to follow in practice and may sometimes require some creative editing.