Inverse problem methods deal with the evaluation of the causal factors that result on a set of measurements or observations. Inverse problems found in nuclear reactors involve non-linear and coupled physical phenomena, making the causation effects complicated to de assessed. Furthermore, the extent of the experimental data collected is limited and this data is subjected to experimental noise. In the following paper, a method for solving inverse problems in nuclear reactors with coupled physical phenomena is developed. In the proposed approach, the inverse problem is solved through the minimization of a performance function. The minimization of this performance function is achieved with a preconditioned gradient descendent method. The generalized gradient of the performance function is obtained using the adjoint of the multiphysics equations of the system. Furthermore, for reducing the sensitivity to noise of the inverse problem, a preconditioner based in a Kalman Filter is developed. As an example, the methodology is applied for solving the inverse problem of finding the heat flux in the wall of a natural convection experiment.