We address some limitations in qualitative and quantitative analysis of time-lapse data due to underlying uncertainties in the flow of reservoir fluids. We explore how seismic signals may evolve over time by developing synthetic time-lapse data based on a variety of flow scenarios including radial flow, multi-phase flow and flows with unstable mobility ratios. Model predictions of water saturation and effective stress are used to construct synthetic impedance profiles. Using an idealized model of water-flood type production we show that qualitative analysis can be limited by the signal complexities resulting from flows in which both water saturation and effective stress change gradually, such as may occur when the injected water and the displaced oil are not separated by a sharp interface. A two-layer reservoir flow model is developed with which the limitations of quantitative analysis are explored. If a quantitative inversion technique is error free the resolution limits and signal noise in the seismic data allow only spatially averaged reservoir properties to be known. In some reservoirs undergoing water flood, volumetrically significant movement of the injected water may occur on length scales below that of the resolution and noise threshold the seismic data allows. In such a reservoir the inversion technique gives good quality information about the average location of the injected water but does not provide information about how all the injected water is moving. Specifically the possibility of thin high permeability layering carrying a significant fraction of the injected water cannot be eliminated. We show that this can introduce great uncertainty in estimates of reservoir performance.