It is widely believed that the cosmological redshift is not a Doppler shift. However, Bunn & Hogg have recently pointed out that to solve this problem properly, one has to transport parallelly the velocity four-vector of a distant galaxy to the observer’s position. Performing such a transport along the null geodesic of photons arriving from the galaxy, they found that the cosmological redshift is purely kinematic. Here we argue that one should rather transport the velocity four-vector along the geodesic connecting the points of intersection of the world-lines of the galaxy and the observer with the hypersurface of constant cosmic time. We find that the resulting relation between the transported velocity and the redshift of arriving photons is not given by a relativistic Doppler formula. Instead, for small redshifts it coincides with the well-known non-relativistic decomposition of the redshift into a Doppler (kinematic) component and a gravitational one. We perform such a decomposition for arbitrary large redshifts and derive a formula for the kinematic component of the cosmological redshift, valid for any Friedman–Lemaître–Robertson–Walker (FLRW) cosmology. In particular, in a universe with Ωm= 0.24 and ΩΛ= 0.76, a quasar at a redshift 6, at the time of emission of photons reaching us today had the recession velocity v= 0.997c. This can be contrasted with v= 0.96c, had the redshift been entirely kinematic. Thus, for recession velocities of such high-redshift sources, the effect of deceleration of the early Universe clearly prevails over the effect of its relatively recent acceleration. Last but not the least, we show that the so-called proper recession velocities of galaxies, commonly used in cosmology, are in fact radial components of the galaxies’ four-velocity vectors. As such, they can indeed attain superluminal values, but should not be regarded as real velocities.