This article from David Cyranoski at Nature News is an excellent behind-the-scenes look at all the problems with the "STAP" stem-cell work, now retracted and apparently without any foundation at all. There were indeed problems with all of it from the start, and one of the key questions is whether these things could have been caught:
The committee was more vexed by instances of manipulated and duplicated images in the STAP papers. Obokata had spliced together gel lanes from different experiments to appear as one. And she had used an image of cells in a teratoma — a tumorous growth that includes multiple types of tissue — that had also appeared in her PhD dissertation. The captions indicated that the image was being used to represent different types of cell in each case. The committee judged that in both instances, although she might not have intended to mislead, she should have been “aware of the danger” and therefore found her guilty of misconduct. Obokata claimed that they were mistakes and has denied wrongdoing. . .
. . .Philip Campbell, editor-in-chief of Nature, says: “We have concluded that we and the referees could not have detected the problems that fatally undermined the papers.” But scientists and publishers say that catching even the less egregious mistakes raises alarm bells that, on further investigation, can lead to more serious problems being discovered.
Many say that the tests should be carried out on all papers. Christopher says that it takes about one-third of her working week to check all accepted manuscripts for the four journals published by EMBO Press. At Nature and the Nature research journals, papers are subjected to random spot-checking of images during the production process. Alice Henchley, a spokeswoman for Nature, says that the journal does not check the images in all papers because of limitations in resources, and that the STAP papers were not checked. But she adds that as one outcome of this episode, editors “have decided to increase the number of checks that we undertake on Nature’s papers. The exact number or proportion of papers that will be checked is still being decided.”
A complication is that some of the common image manipulations (splicing gel lanes, for example) are done in honest attempts to present the data more clearly, or just to save space in a figure. My guess is that admitting this up front, along with submitting copies of the original figures to the editors (and for inclusion in the Supplementary Material?) would help to clear that up. The article raises another good point - that editors are actually worried about confronting every example of image manipulation that they see, for fear of raising the competence of the average image manipulator. There's an evolutionary-arms-race aspect to all this that can't be ignored.
In the end, one gets the impression that Nature's editorial staff (a separate organization from the News people) very much regret ever having accepted the work, as well they might. Opinion seems divided about whether they could have caught the problems with the papers themselves - this was one of those cases where a number of reputable co-authors, at reputable institutions, all screwed up simultaneously when it came to cross-checking and verification. What remains is a portrait of how eager people can be to send in groundbreaking results for publication, and how eager editors can be to publish it. Neither of those are going to change any time soon, are they?
Update: from the comments, see also this timeline of events for a look at the whole story.