This paper studies the use of artificial evolution to automate the design of a satellite passive thermal control system. This type of adaptation often requires the use of computer simulations to evaluate fitness of a large number of candidate solutions. Simulations are required to be expedient and accurate so that solutions can be successfully transferred to reality. We explore a design process that involves three steps. On a first step candidate solutions (implemented as surface paint tiling patterns) are tested using a FEM model and ranked according to their quality to meet mission temperature requirements. On a second step the best individual is implemented as a real physical satellite mockup and tested inside a vacuum chamber, having light sources imitating the effect of solar light. On a third step the simulation model is adapted with data obtained during the real evaluation. These updated models can be further employed for continuing genetic search. Current differences between our simulation and our real physical setup are in the order of 1.45 K mean squared error for faces pointing toward the light source and 2.4 K mean squared errors for shadowed faces. We found that evolved tiling patterns can be 5 K below engineered patterns and 8 K below using unpainted aluminum satellite surfaces. (C) 2016 Elsevier Ltd. All rights reserved.