Noise plays a fundamental role in connection with theoretical limits in the field of inverse thermal problems, since the solution accuracy for such inverse problems is decisively influenced by noise-induced errors in temperature measurements. There are eight standard assumptions proposed by Beck and Arnold (1977), Beck (1979), Beck et al. (1985) and Özisik and Orlande (1999) for the statistical description of measurement errors (noise) which should be fulfilled in order to adequately solve an inverse heat conduction problem. Therefore, it is of utmost importance for thermography measurements that the real infrared camera properties (and especially the noise performance) are well characterized and their influence on the measurement results is known and considered. Within the scope of this work three different mid-wavelength infrared cameras (based on indium antimonide and cadmium mercury telluride detectors) from various manufacturers were characterized with respect to the above mentioned prerequisites by means of the so called photon transfer technique. One of the results show that the tested IR quantum detectors are mostly photon-noise limited and therefore subject to a pronounced Poisson noise characteristic, i.e. the noise signal level (signal variance) depends on the signal intensity. This in turn violates one of the basic prerequisites of most inverse heat conduction approaches, which are relying on an additive noise model with constant variance. In addition, the camera gain constants for two camera models were determined using the photon transfer technique. They are 433 photons/DN or 281 photons/DN and are a measure of the sensitivity of the camera, i.e. the number of incident photons that result in a measurable signal change in the thermographic image.