Elastic inverse-scattering theory has been extended for fluid discrimination using the time-lapse seismic data. The fluid factor, shear modulus, and density are used to parameterize the reference medium and the monitoring medium, and the fluid factor works as the hydrocarbon indicator. The baseline medium is, in the conception of elastic scattering theory, the reference medium, and the monitoring medium is corresponding to the perturbed medium. The difference in the earth properties between the monitoring medium and the baseline medium is taken as the variation in the properties between the reference medium and perturbed medium. The baseline and monitoring data correspond to the background wavefields and measured full fields, respectively. And the variation between the baseline data and monitoring data is taken as the scattered wavefields. Under the above hypothesis, we derived a linearized and qualitative approximation of the reflectivity variation in terms of the changes of fluid factor, shear modulus, and density with the perturbation theory. Incorporating the effect of the wavelet into the reflectivity approximation as the forward solver, we determined a practical prestack inversion approach in a Bayesian scheme to estimate the fluid factor, shear modulus, and density changes directly with the time-lapse seismic data. We evaluated the examples revealing that the proposed approach rendered the estimation of the fluid factor, shear modulus, and density changes stably, even with moderate noise.