The root mean square (RMS) difference of time-lapse seismic amplitudes is routinely used to identify the substituted fluid type in a reservoir during oil field production and recovery. By a time-lapse seismic method, we study the effects of fluid substitution in a physical model, which is an analogy of the three-dimensional inhomogeneous reservoir. For a weak inhomogeneous medium, gas/oil substitution results in positive anomalies in the reservoir layers, and negative anomalies below the bottom of the reservoir layers; while water/oil substitution causes only weak variations in the reservoir layers, but positive anomalies below the bottom of the reservoir layers. For the strong inhomogeneous medium, no matter what kind of fluid substitution (gas/oil or water/oil), there are significant anomalies in seismic amplitude difference attributes both in and below the reservoir layers. Therefore, for weak inhomogeneous media, such as tight sandstone or thin interbedded layers, the RMS amplitude difference attributes can be used to monitor fluid changes and predict the drilling direction; for inhomogeneous medium with karst carves or fractures, it is difficult to accurately determine the distribution of fluids with the RMS amplitude difference attributes.