Mirroring of reference images comes into play in the batch file multirefalign.rct. If the parameter [mirror-option] is set to 1, the particle images will be compared to both the original and mirrored versions of the reference images. I couldn't find an obvious real-life case where particles were flipped by exactly 180 degrees, but a pseudo-twofold in the grid plane caused some confusion. Namely, particles we incorrectly assigned to the flipped version, and during the 3D reconstruction step, unnecessarily flipped particles were combined with the unflipped particles.
If the parameter [mirror-option] is kept at 0 (the default), and you do have 180-degree-flipped particles, the flipped particles will simply form a distinct class, and separate reconstructions will be computed downstream for each class. Since you will typically have multiple classes reconstructed, this inconvenience wouldn't be big.
If you do set the [mirror-option] as 1, there are two checks in the protocol. First, in viewaverages.rct, if the batch file detects Zerodegr/Multiref2/goodsel***-flip and Zerodegr/Multiref2/goodsel***-noflip files, it will generate separate averages. With higher signal-to-noise than the individual images, these averages may help decide if they flipped and unflipped versions are truly the same. The second check is that, in the 3D reconstruction step bpclass.rct, you can set the parameter [flipflop-yn] to 1, and separate reconstructions will be computed for the flipped and unflipped particle images, in addition to the combined reconstruction. If the particles were truly flipped on the grid by 180 degrees, the corresponding reconstructions should not need realignment (since the Euler angles will account for the flip). On the other hand, if the particles were unnecessarily flipped (as was the case with my pseudo-twofold), then the subvolumes will be 180 degrees misaligned.
For what it's worth, using phantom data flipped by exactly 180 degrees while checking for mirrored references seemed to work.
Source: flipflop.htm Page updated: 2012/12/18 Tapu Shaikh