To mitigate global warming, it is necessary to emit less CO2 into the atmosphere and the Carbon dioxide Capture and Storage (CCS) attracts attention these days as one of the solutions against the problem. Off Tomakomai coast, Japan, a CCS project has been demonstrated since 2016. However, there may be a risk of CO2 leakage and consequent seepage from the seafloor, even if the probability of such an event is almost nil. In this research, we assumed that CO2 seeps from multiple points on the seafloor and aimed at estimating the seepage locations, time and fluxes, by using CO2 concentration data observed by several sensors set on the seafloor. We adopted the adjoint marginal sensitivity method, which is a probabilistic time-backward method: an adjoint location probability is released by each sensor and spreads in the time-backward direction. The adjoint location probabilities are used to estimate the seepage fluxes. We also combined the least squares method with the adjoint marginal sensitivity method to estimate the seepage fluxes. We considered that CO2 seeps from 2 points in 2-dimensional horizontal domains as test calculations with changing seepage flux ratios, such as 1:1, 1:0.1 or 1:0.01.