You drive the heater and look at the thermistor temperature. I'm a huge fan of step responses for this job because the high frequency info is all there in the first ~ 3x the rise time. (For a 1-pole response the rise time is 2.2 time constants.) In your situation you'd want to do it at a few different drive levels, because as Tim says, the gain and phase will be different for the two heat sources.
Putting one thermistor by the transistor and another by the resistor will also allow you to null out the effect of the resulting temperature gradient at the position of your sample by changing the relative gain applied to the two. The simplest way is to wire them in series, but then the nulling trick only works at one temperature. (I talk about this in my thermal control chapter someplace.) You can make the null position more stable with respect to the set point by linearizing the thermistors independently and then forming the overall error signal using a linear combination.
Cheers
Phil Hobbs