Modelling of non-stationary signals by time-dependent ARMA model is investigated in this paper. It is shown that a meaningful classification of general time-dependent ARMA model is possible according to time functions of the system poles. Complex exponential, AM, FM, etc. models are shown to be special cases of time-dependent ARMA model. This paper presents a unified approach of parameter estimation of various non-stationary models. The approach provides better flexibility and accuracy in modelling.