We predict sub-skin temperature by using temperature distribution of the skin surface during cryo-anesthesia for the first time. Temperature responses at the area of the process (~10 mm) and the surroundings (~20 mm) are obtained from a patchable temperature sensor array during a non-contact two-phase CO 2 jet cooling. Based on the surface temperature responses, a recurrent neural network (RNN) is trained to predict thermal properties of a few polymer samples with different properties. By using the trained RNN, thermal conductivity and diffusivity of a porcine skin are predicted as 0.155 W/mK and 0.122 mm 2 /s. Then, another RNN is constructed to predict a sub-surface temperature response, which is trained by the temperature at 1-mm depth from the surface using a thermocouple inserted inside the polymer samples. As a result, the subsurface temperature response of porcine skin is predicted with 99.2% accuracy.