Spectral occupation modeling is a key aspect for an adequate prediction of spectral opportunities, which results in a more efficient use of the radio spectrum.The objective of this work is to present a comparative evaluation of the performance of four time series models: AR, MA, ARMA and SARIMA, in the modeling of spectral occupation for cognitive radio networks.The results achieved show that the MA model presents the best relationship between a high precision in the modeling of spectral occupation and a low level of computational cost.