I've been saying for a long time that airport screening for weapons on airplanes may be an insolvable problem and that we need to consider alternatives.  Now research from the University at Buffalo and Georgia State University suggest we may now know why it is such a difficult problem:

Screeners at airport security checkpoints perform an important task in which they search for objects that belong to threat categories in complex X-ray images. New research by cognitive psychologists at UB and Georgia State University explores the cognitive processes that underlie screening, suggests limits on those processes and has implications for the training and evaluation of screeners in the field.

The findings were published in the article "Specific-Token Effects in Screening Tasks: Possible Implications for Aviation Security" in the November-December issue of the American Psychological Association's Journal of Experimental Psychology: Learning, Memory and Cognition (Vol. 31, No. 6) and in the article "Visual Search and the Collapse of Categorization," in the November-December issue of the APA's Journal of Experimental Psychology: General (Vol. 134, No. 4).

...

The problem addressed by the research, Smith says, "is that screeners must be able to bring 'category-level' knowledge to their search for targets. That is, they must search for guns and knives generally, not for specific Beretta guns or Bowie knives they have been trained to recognize. Yet it has been unknown how the processes of categorization stand up to visual complexity or why they fail facing it. We filled this research gap by creating a visual-search and categorization paradigm in which participants searched for members of target categories in complex displays."

...

The reliance on familiarity might also have implications for the training and evaluation of screeners in the field for the following reason: The "Threat Image Projection" (TIP) system provides an important potential means of assessing performance at security checkpoints. It includes a library of test images that can be digitally injected into the X-ray image of a bag as it is presented to the screener. This lets screeners' ongoing target-detection performance be evaluated, provides ongoing training and maintains screener vigilance. Yet, a strategy of using a fixed library of test images risks the specific-token effects that Smith and his colleagues observed. Estimates of screener performance might be inflated by these familiarity effects, compared to the detection levels that would be observed for real threats that will be from outside the library and unfamiliar. A more conservative and truer estimate of screener performance would result if one arranged testing so that targets essentially never repeated, so that familiarity could never develop and so that screeners were forced to rely on category general knowledge and strategies.

Smith notes that it is to the Transportation Security Administration's credit that it has actively responded to this specific-token effect by increasing the size of TIP libraries and by planning periodically to infuse new targets. These changes have the potential to keep TIP as a viable and accurate indicator of detection levels in the security system. These steps also show the promise of cognitive scientists and sponsoring agencies cooperating toward solving important problems. 

Unfortunately, knowing why it is a difficult problem also means everyone knows how to make it difficult for the screeners.  We need to consider the alternatives.