This app note has two best circuits: first, Figure 6 shows a wideband thermal RMS-to-DC converter. This circuit uses the old servo-multiplier (or light-bulb multiplier) technique of using a feedback loop to force two matched parameters (in this case, the heat produced in the two composite thermistors) to have equal value. We'll see this circuit again. The flowmeter (Figure 8) and anemometer (Figure 11) are related tricks.
The second celebrity circuit is on the final page: Jim's obsession with the HP200 makes an early appearance. Figure 12 is an op-amp recreation of the Hewlett-Packard HP200 audio oscillator circuit (and the references include William Hewlett's thesis). While the flowmeter and anemometer are useful applications, I wonder if they weren't just a set up so that he could justify spending time at work playing with Hewlett's light-bulb circuit.
The other applications are also interesting and instructive. Figure 1 shows an interesting temperature controller (but the LT3525A is now discontinued). The explanation includes a simplified discussion of the various delays and time constants in a heat flow problem, but the full story is much worse: it's not just a multiple-time-constant system, it's a diffusion-equation system! Also, his discussion of insulation needs a caveat about the trade-off: you want to keep the losses small to save power, but large enough to allow the system to quickly recover from an over-temperature condition. Regardless, using the 50-ohm resistor and switch to check the loop response is a great idea, but the loop response needs to be checked for positive steps (when the heater heats) and negative steps (when the losses cool). He only shows the former. (Also, I don't think he needs the 100M resistor in the integrator, but that's another discussion.)
The circuit in Figure 4 uses a CA3096 array to thermally stabilize the feedback transistor in a log amp. This neat trick is also used to thermally stabilize a VCO in National Semiconductor's App Note 286. I'll comment here: Yikes! That LM301A has a huge compensation capacitor! A 33nF capacitor gives a unity-gain frequency of 1 kHz and a slew rate of 0.5 V/ms (yes, volts per millisecond). Replace the LM301A and 33nF with an integrator, and I think you'll get better (and more repeatable) performance. Looking back at Figure 1, is Jim afraid of integrators? I'm guessing Bob Widlar didn't consult on this circuit.
Best quote (page AN5-1): "The close relationship between temperature and electronic devices is the source of more design headaches than any other consideration." Word.