Several works have concluded that the number of cells per frame of VBR video follows a gamma distribution. But few model meets this character. In this paper we propose a novel video model

a first-order autoregressive （AR） gamma sequence

in which random variables are marginally distributed as gamma variables. We give the definition of GAR model and show some characters of GAR（1）. The parameter estimation method is studied too. Simulation results show that our model can reflect the characters of VBR video better than the former models. It is more easy and accurate to create simulation data sequences using this new model.