The paper investigates the possibility of applying a heuristic algorithm to solve an inverse heat conduction problem encountered in Temperature Oscillation IR Thermography method. The idea of TOIRT [24] is to infer about convection coefficient in given point h(x,y) from phase delay φ(x,y) of the temperature signal measured by means of thermography on the opposite side of heat conducting wall in steady-periodic conditions. Since the exact solution for inverse 3D heat conduction problem defined by TOIRT method is not available, there is a need to employ some kind of optimisation scheme to find the distribution h(x,y). In [22] it has been done by iterative solution of associated direct problem with finite difference method and updating the distribution h(x,y) in every iteration according to relative error between measured and computed values of φ(x,y) on the surface of the conducting wall. This paper presents an alternative approach based on symbolic regression with Genetic Algorithm to find an explicit formula relating amplitude as well as phase shift angle of temperature signal and the value of convection coefficient. Proposed procedure involves an offline training of the model with data previously generated with Finite Element Method which assures that time-consuming part of the calculation is carried out once and since then obtained explicit formula may be used in fast and straightforward manner. The results show that proposed approach gives solution with at least comparable accuracy to that obtained in [22], but requires careful selection of training data mainly in terms of their diversity and values of gradient of h(x,y).