Phobos’ parallel grooves, which are such a striking feature of its surface, have attracted interest since their discovery on Viking images in 1976 (Veverka and Duxbury, 1977. J. Geophys. Res. 82, 4213–4223.), but their origin is still in dispute today. The great increase in knowledge of Phobos’ surface features effected by images from the High Resolution Stereo Camera (HRSC) onboard the E.S.A. Mars Express spacecraft has clearly demonstrated that only one hypothesis can seriously be upheld: that the grooves are chains of secondary impacts resulting from primary impact events on Mars (Murray and Iliffe, 2011. Martian Geomorphology. Geological Society, London, Special Publications, 356, 21–41. DOI: 10.1144/SP356.3). But even this hypothesis has recently been questioned from a ballistic standpoint (Ramsley and Head, 2014. Planet. Space Sci. 75, 69–95.) mainly using estimated data for the grooves themselves. In the present paper we present summaries of extensive new measurements of groove sizes, pit separations, groove family parameters and geographical distribution. We also re-examine their unique characteristics, and present a review of past ideas. But we concentrate on extending and refining the work of Ramsley & Head using the new measurements, and the much-improved geodetic data from Mars Express (Willner et al., 2014. Planet. Space Sci. (2014)). We find that the total mass of Mars ejecta required to form all the observed grooves on Phobos is between 2.0×109 and 2.7×1010kg, and that the total mass of impact ejecta from all Mars craters between 19 and 384km diameter likely to hit Phobos (in its present orbit) at sufficient velocity to form all the grooves is one or two orders of magnitude greater than this. The larger available ejecta mass from Mars is due to several factors, including the fact that Phobos is known to have orbited Mars at a greater distance at the time when the grooves were formed. A new rigorous celestial mechanical analysis of the transfer of ejecta from Mars to Phobos is presented, including N-body calculations that model not only the gravitational interaction of Mars, but also of Phobos and up to several hundred individual ejecta particles. This allows us to model the possible source areas on Mars of all observed groove families, and to conclude that more than one groove family may have been formed by different batches of ejecta from the same Mars impact event, and that the total number of Mars impact events responsible for creating the grooves may have been less than 10. The N-body calculations also allow us to model the complex gravitational interactions between Mars ejecta particles, and thus to assess the contrast in groove morphometry between those families centred on the sub-Mars and anti-Mars hemispheres.