The possibility of using time-varying parameter models in the context of error correction models is studied empirically. As an application, a money demand relationship (M1) for Venezuela is estimated from 1983 to 1994 within a cointegrated VAR framework. First, the stochastic properties of the series are analysed, studying each corresponding order of integration. Second, the existence of a long-run stable relation between the variables involved has been investigated, and then the cointegration relation and the short-run adjustment mechanism estimated. As both relations are identified in the context of constant parameters a stability analysis is performed. Finally, the technique of Kalman filtering is used to estimate a model that permits the short-run parameters to vary, while the parameters of the long-run relation are kept constant.