Agree with phage...a filter membrane is not meant to separate proteins from other proteins. I've had monodisperse monomeric GFP (28kDa) be largely retained by a 100kDa filter.
Cation or anion exchange is your best bet. Or use a different affinity tag. Or it's possible your 26kDa protein is a proteolytic breakdown product of your target protein, in which case throw in some protease inhibitors (and use clean, fresh filtered buffers, work at 4C or on ice, etc).
Lots of people who start working with proteins are told all proteins can be purified with the same protocol, i.e. with just an affinity column followed by gel filtration. This is very commonly not the case. Expect to put in a little effort to purify quality protein if it's anything other than a well behaved, very well expressed, super stable and soluble protein (which most aren't ). Don't get caught up being protocol oriented when working with proteins. I see too many people doing it these days and their work suffers because of it.
EDIT: in grad school I had a protein with an intraprotein (internal) crosslink. The crosslinked protein traveled about 5kDa faster on a gel than the non-crosslinked. In other words, if you can convince yourself both bands on your gel are your target protein (in-gel digest + MS/MS will do the trick), consider in addition to proteolytic cleavage other post-translational modifications that could alter electrophoretic mobility such as internal crosslinks.
Edited by labtastic, 26 August 2014 - 09:56 AM.