Steady-case thermal analysis plays an important role in dimensioning thermal control systems for spacecrafts and aircrafts. Usually a trial and error approach is used based on engineering judgement and experience. When thermal models become complex or there are conflicting thermal requirements, however, it becomes harder for an engineer to gain insight as to which design decisions will lead to better results. Numerical optimization, on the other hand, could provide a more robust approach for the thermal design of complex spacecraft or aircraft models. In this paper, we suggest a gradient-based multidisciplinary optimization of thermal models where the coupled derivatives of the multidisciplinary system are obtained with the adjoint method. We show that in the case of steady-state thermal analysis, there is an analytic solution of a partial derivatives of implicit heat-transfer equation that can be used to derive total derivatives of the system. We present a practical application of this method by solving a small interplanetary spacecraft thermal optimization problem consisting of one objective, 15 design variables, and 10 constraints. We found that by using gradient-based optimization with exact derivatives, the best results can be achieved by exploring the design space at multiple initial starting points without major computational overhead.