It is known that the propagation velocity of sound waves in water can vary over time. For a 3D seismic survey, if data are acquired in adjacent lines but at different dates, this implies the same reflection point will be recorded at different times. To take this effect into account in seismic processing, it is necessary to measure the sound velocity in water. I have developed a 3D tomographic method that directly estimates it. It assumes a constant sound velocity for a group of shots belonging to a single sail line. Using a picked water-bottom reflection and an initial depth and velocity model, results good for use in subsequent processing can be obtained by estimating only two parameters: the variation of the propagation velocity and a constant vertical shift of the reflector depth in relation to the initial model. The method was tested with both synthetic and real data. The real data results were validated using two methods. First, I analyzed the histogram of the residuals of the final updated model. Second, I used a specially modified Kirchhoff migration algorithm to migrate the sea bottom. The main advantages of this method are that it takes into account the sea-bottom dips to estimate the velocities and it can be applied to each sail line separately. Also, the inversion is not ill-conditioned provided that data with large enough offsets are used. As a result, the method is simple to apply.