In the delayed choice quantum eraser experiment of 2000 (DCQE http://en.wikipedia.org/wiki/Delayed_choice_quantum_eraser) which I have attached a simplified figure for, there is the use of a downconversion crystal which converts pump photons into two photons of half the energy, signal and idler. In the DCQE, the pump is fed through one of two slits which are separated by 0.7mm, so there is the possibility that each photon may have travelled through either slit prior to downconversion. Since the two possible downconversion processes take place within the same crystal, they are coherent. Therefore there is the possibility of interference between the two downconversions. The signal photons are fed to a detector where there may be interference observed. The interference there depends upon the "delayed choice" of the experimenter. Some of the idlers are reflected off the beamsplitters and detected, which implies "which path" information and negates the interference at the signal detector. Some idlers transmit through the beamsplitters and are forced to interfere with the other "possible path", which erases the which-path information. This allows interference at the signal detector, and it is noted by setting up coincidence counters between all detectors and post-selecting the idlers that transmitted. Simple, and very similar to Mandel's ZWM from 1991 (but you'll notice the authors did not reference the ZWM). However, Mandel finds interference in the setup he uses without post-selection required. Why does the DCQE protocol require post-selection?