In this paper we examine the core surface flow obtained by an inversion of a continuous model of the geomagnetic field and its temporal variation using the diffusion-less induction equation. The continuous CHAOS model is derived from satellite data up to spherical harmonic degree 14 and covers the period 1999 to 2006. The CHAOS secular variation, when downward continued to the core surface, shows stripe-like features, which can be attributed to spherical harmonic degree 12 and higher. These contributions are removed by applying a tapering method, and the resulting tapered model is then inverted for the core surface flow. Satellite-based field models have a high spatial resolution; however, their temporal resolution is limited. In order to enhance the temporal resolution of the flow, we additionally constrain the flow to fit the secular variation from ground-based observatory data. A range of solutions, subject to different constraints, are computed, two flow hypotheses being considered: purely toroidal flow and tangentially geostrophic flow. We show that both flow types provide similar results; however, the purely toroidal flow provides a better fit to the secular variation in the equatorial region than the tangentially geostrophic flow. We then analyze the residuals between observed secular variation and its predictions from the flow. We note larger residuals for the tangentially geostrophic flow, where strong radial secular variation and a weak radial field are observed. Although diffusive effects cannot be ruled out as a potential source of the mismatch, we attribute the larger residuals to be caused by a flawed estimation of the poloidal flow. We also seek to relate temporal changes in the fluid flow to the geomagnetic jerk which occurred at the beginning of 2003. It is found that this geomagnetic jerk coincides with variations in the zonal flow components of both flow types, suggesting a possible link to torsional oscillations. However, we argue that other flow components are important to obtain a detailed fit of the observed secular variation. Finally, we compare observed changes in the length-of-day and the predictions from the flow solutions. (C) 2008 Published by Elsevier B.V.