As a fairly finicky photometrist, my reaction is that either the star is constant at the couple-percent level and the differences you point out (which are quite small) arise mainly from modest observation/ reduction effects, or that the star could be very slightly variable. Among the published results for the star collected at Mermilliod's Web site (http://obswww.unige.ch/gcpd/gcpd.html), here's what I see: UBV: _none_ of the UBV data can be relied upon except the 'final' Johnson values from the LPL monograph: V=2.98, B-V=1.40. These results probably include data from the original tube and filters, so in effect cannot be argued with, since they more-or-less define the system. However, it is clear that Johnson's V magnitude zero-point was not fixed to any better than 0.02-0.03 mag., so there is plenty of room for slop by +/- a few percent. Geneva: the Geneva V zero-point has systematic error that is a function of Right Ascension. Maybe their value for epsilon Gem is too faint because it is in a low spot in this relation. uvby: the values to rely on here are from Erik Olsen and collaborators. He used the 1966 Johnson V magnitudes as his reference, and because of the large number of observations (many thousands for the standard stars) the V zero-point is accurate. I have tested Olsen's V directly against Landolt equatorial standards and conclude that his zero-point is within a few millimag of the Landolt system with very small (~0.007 mag.) scatter star-by-star. Thus his values of V=3.002 and 3.008 are probably reliable. WBVR: the other place I'd look is the very reliable survey of the 13,000 brightest northern stars by Kornilov et al. Internal errors are commendably small (7-9 millimag), but there are systematic errors in both V and B-V (at least) such that for red stars V is too faint and B-V too red. For epsilon Gem we find V=3.003 and B-V=1.435. From my experience I'd expect 'true' V to be something like 2.99. So from the really reliable photometry, V seems to be something like 2.99 to 3.01. The 2 percent range is still very much within the slop expected from small passband 'funnies' in the reductions, choice of standard stars, etc. From the data at hand you simply can't decide whether the star is variable---it could be utterly constant and the published results could show much wider range. Probably the only way to decide such a case would be new _differential_ observations with several comparison stars. \Brian