Background information:
Binaural Masking Level Difference (BLMD) [1] is a phenomenon that is observed at low frequencies when a masker and probe have specific time relationships with respect to each other (a "masker" is the signal doing the masking, a "probe" the signal that is being masked).

The classic experiment can be described as follows:

A narrowband noise masker of one critical bandwidth is presented identically at both ears (i.e. the same time signal), while the phase of the masked sinusoidal probe is alternated, i.e. presented in phase, and then out of phase, in the two ears. Conversely, the phase of the masker can be alternated while the phase of the probe stays the same. In both cases, the interaural phase relation for the masker vs. probe is radically changed. The sensitivity of the auditory system to the two changes is not symmetric, however. It is also possible to repeat the experiment with a sinusoidal masker, and narrowband noise as the probe, however the effect, while present, is substantially reduced.

For the full effect to be noticed, this must be done at or below 500 Hz center frequency, although some effect has been noted up to between 2 and 3kHz.

In the case where both masker and probe are the same in both ears, a masking threshold very much like the single-ear masking threshold is observed, i.e. for the example given, the threshold of masking for the tone probe is approximately 5.5 dB, as shown in the literature. In the case where the probe is out of phase (but the masker still in phase), a difference between the signal with and without the probe is easily audible at this level.

The experiment can also be run with the masker being applied in and out of phase, in which case release of the masking threshold is also observed, in some specific cases of up to approximately 20 dB.

Sound examples
he two signal pairs presented here are the classic signals, created directly from MATLAB. Test signals A1 and A2 use in-phase probe tones of 400Hz frequency with probe levels of 9dB and 6dB below the level of the masker, respectively. Test signals B1 and B2 use out-of-phase probe tones with the same frequency and probe levels of 9dB and 6dB below the levels of the masker, respectively. The masker noise is presented in-phase in all cases, despite the sometimes startling spatial effects arising from masking release.

Test Signal A1 Probe in-phase, probe level 9dB below masker

Test Signal A2 Probe in-phase, probe level 6dB below masker

Test Signal B1 Probe out-of-phase, probe level 9dB below masker

Test Signal B2 Probe out-of-phase,  probe level 6dB below masker

When listening to signal pair A1/A2, increasing the probe level does not lead to an audible difference in the perceived sound since the probe is masked in both cases. This is in line with classic monaural psychoacoustic observations which assume a masking threshold of ca. 5.5-6dB below probe level for this case. On the other hand, when listening to test signal B1 and comparing to test signal A1 (i.e. switching from an in-phase to an out-of-phase probe), an audible difference in sound impression is perceived due to the reduction of the masking threshold by the BMLD effect. This effect is even more pronounced when listening to test signal B2, i.e. with a level of 6dB below the masker.

While these signals do not directly mimic most coding artifacts (although they do demonstrate the same effect as observed in certain well-known test items, such as Suzanne Vega / Tom's Diner in mono vs. stereo at the first stereo test), they do provide a good example of the kind of artifacts to listen for, and one that has proven good at sensitizing listeners to more subtle coder-induced artifacts involving BMLD.

In order to hear imaging artifacts in general, the listener must stop, at least temporarily, focusing on the usual range of artifacts, and try to allow the stereo signal to construct a soundstage, and then listen to artifacts. More specifically, one should attempt to position of things in the soundstage, noting both omissions and commissions, i.e. new additions to the soundstage that are not present in the reference. Again, both errors of Omission and COmission must be noted, as either can happen.

It is worthwhile to note that while the classic BMLD does not operate above about 2kHz, there is a similar effect based on the signal envelope that can create audible problems, for instance in certain recordings of a Harpsichord, an effect that can be noticed as a "dirty surround" at frequencies above 3kHz. The occurance of such problems is very signal-specific, but again related to the spatial perception of the signal and any new or missing parts in the spatial response.

One interesting case that has been observed is in fact a "missing" artifact, where in rock cut called "Dorita" by Lou Reed, the stereophonic presentation of a coded signal appeared to be missing high frequencies, while either channel of the signal, presented monophonically (either in 1 or two speakers) did not. This effect was eventually correlated with the high-frequency envelope of the signal in the two channels. When the high-frequency envelope of the two channels was effectively randomized, (it being highly correlated, but delayed a bit, in the original) the auditory system appeared to understand the high-frequency sounds as not part of the guitar "auditory object". When the high-frequency envelope was corrected, the high frequencies returned to the sound of the guitar. Bear in mind that through this experience, the high frequencies sounded normal in a monophonic presentation, be it of left, right, or summed L+R channels.  

References:
[1] J. Blauert, "Spatial Hearing", MIT Press, 1983