A number of seismic processing methods, including velocity analysis (Sheriff and Geldart, 1999), make the assumption that recorded waves are primaries - that they have scattered only once (the Born approximation). Multiples then represent a source of coherent noise and must be suppressed to avoid artefacts. There are different approaches to mitigate free surface multiples (see Dragoset et al. (2010) for an overview), but internal multiples still pose a problem and usually cannot be removed without high computational cost or knowledge of the medium. Recently, Marchenko redatuming has been developed to image a medium in the presence of internal multiples (Wapenaar et al., 2014). Using Marchenko redatuming in combination with convolutional interferometry, Meles et al. (2016) have developed a method which allows the construction of a primaries-only data set from existing seismic reflection data and an initial velocity model. The method was proposed for the acoustic case and appears to be robust with respect to even huge inaccuracies in the employed velocity model. In this paper we investigate the impact of such primaries-only data on a simple velocity analysis workflow, as opposed to using the full data set with multiples. We use semblance analysis (Sheriff and Geldart, 1999) and compare the results obtained with three different data sets: the full reflection data with multiples, primaries data calculated with prior knowledge of the subsurface, and primaries data calculated with an entirely incorrect constant velocity model. We then use the velocity models that we construct to perform reverse time migration (RTM) of each of the data sets. We find that the velocities found are robust with respect to errors in the initial model used for Marchenko redatuming, and the method produces good results if non-hyperbolic moveout effects are avoided. @en