I like the antibody from Cell Signaling Technology (Ezh2 (D2C9) XP® Rabbit mAb #5246). It gives a very similar result as an antibody from Novocastra (NCL-L-EZH2; http://www.proteinatlas.org/ENSG00000106462/tissue). Since they seem to be 2 independent clones (one from mouse, one from rabbit) depicting very similar staining patterns on human FFPE tissues, I dare to extrapolate and claim that both reliably bind EZH2.

The mouse anti-EZH2 antibody (612667, BD Transduction Laboratories)  appears to follow the same staining pattern but I can't comment on this with absolute certainty.

Generally, an antibody should be used whose sating pattern can be confirmed by another antibody (preferentially against a different epitope of the same protein). This is often difficult to find but for EZH2 the 3 above mentioned antibodies seem to fall in this category.