Other 17% (presumably Edwards, Dodd, Richardson, and other presidential candidates who have suspended campaigns but remain on the Texas ballot) and Undecided 4%.
This outfit -- People Calling People/Texas Voyager (.pdf, scroll to bottom) -- used an automated interactive call to 408 respondents, self-declared Democrats who have already voted in the March 4 primary. The MoE is 4.85%, and calls were conducted between 6 and 9 p.m. on the evenings of February 26 and 27.
Neither Kuffner nor BOR, the real gurus for these sorts of numbers, appear to have blogged this yet so opinions as to validity, integrity and so on I'll leave to them and similar experts as to whether or not this is good data.
It seems to contradict the supposition that the "surge x 10" of early voting is driven by the Obama campaign. I suppose we will have to wait and see.
Update: That 17% 'other' -- as well as the 4% undecided -- calls into question the accuracy of an exit poll, particularly one in which the participants declare as having voted to an auto-dialer. K-T at BOR mentions that the poll doesn't appear in the Pollster.com list, or the Real Clear Politics aggregate.
Update (2/29): Kuffner -- with the assist from John Zogby -- makes it make a little more sense.