In addition to points made in other answers, and strengthening Dave Clarke's first sentence: not only the best, but even "medium-good" journals receive far more good submissions than they could possibly publish ... at least within the traditional "page limit" constraints, and possibly within some tacit constraint about appearing sufficiently selective. (That is, even if all the papers were quite good, a journal that publishes 10,000 pages a year will be suspect...)
Thus, the job of "editor" tends to degenerate into trying to find reasons to reject.
There simply are not enough "prestigious" journals to publish all the (good!) work done by all the good young people... who now can typeset things much faster and get them submitted. (In contrast, pre-TeX, pre-internet, getting something sufficiently presentable to send to a journal was much more of a hassle... Also, the number of researchers has grown much faster than the number of good journals, at least in my field, mathematics.)
Returning to the literal question: as others have said, indeed, try to be constructive and helpful, rather than merely a gatekeeper. At the same time, you cannot count on corrective feedback from editors if you are "too harsh", because they may not realize this, and in fact have incentive not to worry about that exigency, because you are making their job easier.
Also, one should note that in recent times many respectable journals explicitly declare to referees that they, the referees, are not responsible for correctness of an article. It is supposedly the author's responsibility. This is obviously strange, but can be understood when we realize that the gatekeeper function of journals, editors, and referees is at least as important as the scientific function... as disillusioning as that might be. But it does explain situations that are otherwise baffling and counter-intuitive.