I agree with Fretwell. I think that the gfi tester works by allowing a small amount of current to flow from the hot conductor through a resistor to the grounding conductor. So if you have no ground connection at the gfi the tester's ground prong wouldn't have a good return path for the simulated ground fault.
This doesn't mean that the gfi won't work as intended though since it normally compares line current to neutral current, with no need for a ground.
Perhaps you could verify it's correct operation by unscrewing the gfi and, using a separate wire, make a temporary connection to a good remote ground point and then try your tester again?
Shawn.
