A 1K 0.1C thermistor seems like a bad use-case for a wheatstone bridge. With a typical Beta of 3800, a 1K NTC thermistor at 25C will vary about 42 ohms (~4%) per degree of temperature change. In a 5V Wheatstone bridge config with 1K resistors, the +/- 4% change in resistance per degree will make the voltage difference across the terminals of the Wheatstone about +/- 4V. With a 10 bit ADCs on the terminals of the wheatstone, you might sense 0.1% differences, which would correspond to temperature differences in the +/-0.025C range.
But, with a 1K thermistor at 2.5V, you could be heating the thermistor significantly: 2.5V/1000ohm=2.5mA, P=2.5ma*2.5V=6.25mW With one particular 1K thermistor's dissipation constant of 1.7mW/C, it could self-heat by 3.67C, introducing a systematic error 147 times larger than the sensitivity of the wheatstone.
If you need to do this with an arduino, use a larger resistance thermistor and a very small reference voltage to minimize the self heating. And figure out your operating range and what that would mean at the terminals.
The other problem is how you are going to to the +/- differential voltage measurement on the arduino. If you hook the AREF to your wheatstone bridge's reference voltage, half of the wheatstone bridge is essentially useless-- you can do the inactive branch of the wheatstone bridge in math rather than measuring the terminals with ADCs.
Some AVRs can do differential +/- voltage measurements with some ADMUX magic: http://arduino.cc/forum/index.php?topic=147830.0