It's obvious that really dumb questions with little votes can be voted down immediately. But imagine you have a totally average question which, in your opinion, bears no great interest to the general audience of the site, but has been upvoted in excess of 10 times – do you use a downvote to "correct" the score in the direction of your score – even if you might upvote the same question if it had only, say, 1 or 2 upvotes because you think this question is kind of worth 3 upvotes, no more and no less?
The same applies to the other way around as well. Maybe there is a bad question, but not so bad that it would justify 10 downvotes. Even if you would usually give 1 downvote, would you correct the score towards your desired target with an upvote?
Also, you might really like a question – but if it already has 20 upvotes, do you refrain from upvoting it even more? (I'm actually quite sure that scores would be much different if no-one could see votes by others.)
Or do you really only listen to your first impression towards a question when voting, without acknowledging existing votes?
I have the feeling that sometimes I want to "correct" the score of a question with my vote even if it's not an inherently bad question, just a totally overrated one – but I want to be sure that this is acceptable behaviour before I actually go on and deal a downvote to a user (I am of course aware that I lose rep as well by downvoting).
Edit based on first answer: Why do so many questions linger around 1 or 2 upvotes? Isn't that a sign that many users think "yes, this score is ok, but it doesn't deserve more"? I'm not convinced that it really represents the number of users who like a question. For example, I have noticed how community wiki questions tend to have much higher scores – I think this is a sign that people don't want to be so generous with giving reputation to others, but if no reputation is dealt, then they freely decide ...
Edit 2: I also just found this: Why a*a*a*a*a*a cannot be optimized to (a*a*a)*(a*a*a)? , note Steve Jessop's comment on the answer concerning his upvote policy ("more likely to upvote if there are incorrect answers present").