This paper is devoted to set membership parameter estimation for non-linear complex-valued models. In such a context, the error between the measured data and the output model is supposed to be bounded with known prior bounds. The proposed approach is based on set inversion via complex interval analysis. A first contribution of this paper is the extension of logarithm and exponentiel functions to polar intervals. In a second part, it is shown on an engineering application devoted to thermal parameter estimation, that the proposed interval complex representation is useful.