Structure: every comment, in order, quoted
Reproduce each reviewer comment in full, in the order it was made, and answer it directly underneath. Number them by reviewer — Reviewer 1, Comment 1 — so that the editor and each reviewer can find their own points without reading the whole document.
Distinguish the comment from your reply visually. Italics or a different weight for the reviewer’s text, plain text for the response, and the revised manuscript wording quoted where it helps. An editor reading twenty of these a week should be able to follow yours without effort.
Give the location of every change: page and line number, or section heading. "We have added a paragraph on this" sends the reader hunting. "We have added this at p. 7, lines 142–151" does not, and it signals that the change genuinely exists.
Quote the comment, state the change, give the line number. An editor should never have to search for what you did.
Answer the point that was made
Reviewers frequently raise an underlying concern through a specific request. A comment asking for an additional control may really be asking whether an alternative explanation has been excluded. Address the concern as well as the request, and say which you have done.
Where you made the change, say so plainly and show it. Where you made a different change that meets the same concern, explain why your version answers the point — reviewers usually accept an alternative that works, and rarely accept silence.
Where a request is impossible — an experiment requiring an instrument you do not have, data that cannot be collected retrospectively — say that directly, explain why, and offer what you can: a limitation stated in the discussion, an alternative analysis, a narrowed claim.
Disagreeing without losing the reviewer
You are allowed to disagree, and editors expect it occasionally. What matters is that the disagreement is argued from evidence rather than asserted, and that it is courteous — the reviewer will read this letter, usually in the same round.
Acknowledge the point before you answer it. "The reviewer is right that X would strengthen this; we did not take that approach because Y" is a different document from "we disagree." Cite the literature or the data that supports your position, and keep it short.
Where a disagreement is genuinely unresolvable, concede ground somewhere visible — soften an over-strong claim, add the caveat the reviewer is circling. A revision that changed nothing anywhere reads as a refusal to revise, whatever the letter says.
What goes to the editor
Open with a short covering note to the editor: thank them for the opportunity to revise, state in two or three sentences what changed substantively, and confirm that the response letter addresses every comment.
Flag anything the editor specifically needs to know — a comment you have declined and why, a change of authorship, new data added since submission, or an overlap with another paper of yours now in press.
Submit both files: the point-by-point response and the revised manuscript. Where the journal asks for a marked-up version, produce it — and check that a clean version is what goes forward for typesetting, since sending tracked changes into production causes avoidable delay.