Summary A key step in a seismic velocity study involves velocity interpolation, particularly when a true integration between seismic and well velocities is required. Additionally, velocity can be seen as a measurement of the vertical instantaneous velocity field, while stacking derived velocity, even when corrected by the cable length, is an integral measurement of the (mainly) horizontal velocity field. It is of major importance to take into account the different scales of these two kinds of measurements. A 3D geostatistical interpolation of the instantaneous velocities requires the knowledge of the 3D covariance function of the instantaneous velocity field. Currently proposed geostatistical inversion procedures often use a combination of the vertical instantaneous velocity covariance (from sonic velocities) and the stacking derived velocity’s horizontal covariance. This procedure introduces a bias in the estimation process, as the stacking derived velocity covariance may be quite different from the real instantaneous velocity covariance. We show here that horizontal covariance information can be recovered from the covariance of the stacking velocities. The interest of the proposed formula is that it only requires information obtained during a conventional processing of the data. An application of this method to real data is given, along with an estimation of the horizontal covariance length. In order to evaluate the efficiency of this method, tests are currently done using the Touati and Iooss pre-stack traveltimes variance’s method (Iooss 1998), which seems to give estimations of the same order.