Bell Inequalities with Postselection

Experimental tests of Bell inequalities ought to take into account all detection events. If the latter are postselected, and only some of these events are included in the statistical analysis, a Bell inequality may be violated, even by purely classical correlations. The paradoxical properties of Werner states, recently pointed out by Popescu, can be explained as the result of a postselection of the detection events, or, equivalently, as due to the preparation of a new state by means of a nonlocal process.