This paper presents several non-linear models adjusted in order to capture the dynamics of a gypsum kiln. The behavior of this kind of processes is affected by non-linear effects caused by the existence of disturbances and the coupling among several variables. The use of second order Volterra and Hammerstein models as appropriate solutions to describe process dynamics is analyzed. A thorough study of the best model order and structure is performed. Coefficients that best fit real data are also selected. This work aims to obtain a good non-linear model in order to implement a non-linear predictive controller, able to improve the performances of those linear controllers already tested on the plant.