A mathematical model of a rotary cement kiln is presented. Available real data are very limited and also contaminated with considerable noise and disturbance. For detrending the data, two approaches are applied. In the first approach, a low pass digital filter is used to remove the low frequency components of the signal. In the second approach, an integrated random walk model is used. Autoregressive models are constructed on the basis of the detrended data obtained by the respective approaches. The results of these two models are compared by simulating the step responses.