For example, in the following master branch, I need to trash just the commit af5c7bf16e6f04321f966b4231371b21475bc4da, which is the second due to previous rebase:
If rebase is an option, you can rebase and just drop it:
$ git rebase -i 414ceffc^
If rebase is not an option, you can just revert it:
$ git revert af5c7bf16
Despite all the credit the original answers received here, I didn't quite find them to satisfactorily answer the question. If you find yourself in a situation where you need to remove a commit, or a collection of commits from the middle of the history, this is what I suggest:
Here's info on Cherry Picking: What does cherry-picking a commit with git mean?
Here's some on doing it with Tortoise Git (as I just did). It's definitely easier to use a gui utility for these sorts of operations! Cherry pick using TortoiseGit
Rebase or revert are the options. Rebase will actually remove the commit from the history so it will look like that second commit never existed. This will be a problem if you've pushed the master branch out to any other repos. If you try to push after a rebase in this case, git will give you a reject non fast-forward merges error.
Revert is the correct solution when the branch has been shared with other repos. git revert af5c7bf16
will make a new commit that simply reverses the changes that af5c7bf16 introduced. This way the history is not rewritten, you maintain a clear record of the mistake, and other repos will accept the push.
Here's a good way to erase: git rebase -i <commit>^
That takes you to the commit just before the one you want to remove. The interactive editor will show you a list of all the commits back to that point. You can pick, squash, etc. In this case remove the line for the commit you want to erase and save the file. Rebase will finish its work.