Abstract. Venusian canali have many characteristics of terrestrial rivers, notably cutoff meanders, braiding, point bars, and deltas, which required both erosion and sediment transport. This implies that the canali were not formed by construction but rather by thermal or mechanical erosion. We have evaluated the relative importance of these latter two mechanisms, assuming a basaltic substrate and a surface temperature similar to that currently prevailing, -470øC. In order to have thermally eroded the canali, the liquid must have been turbulent and at a temperature above that of the basalt solidus. The most plausible candidates for this liquid are basalt and komatiite lavas. However, at realistic flow rates and extrusion temperatures, flow of basaltic lava is laminar, and therefore basaltic lavas could not have thermally eroded the canali. Although komatiite flow is initially turbulent, the lava will cool in hours to its solidus temperature, whereas it will take months to thermally erode canali. By elimination, only mechanical erosion can adequately explain canali formation. Based on incision and lateral migration rates for terrestrial rivers, it could take from >5 years (in unconsolidated regolith) to 8 x 105 years (in solid basalt) to mechanically eroc[e a typical Venusian canale. These estimates require that the eroding agent had a solidus temperature close to the Venusian surface temperature and that viscosities remained low until solidification. Only halogen-rich, alkali carbonatite and sulfur lavas meet these criteria, and only the former could have been present in sufficient volumes to form the canali. We propose that the canali were mechanically eroded by such carbonatite lavas and that the latter originated from the fusion of anhydrous recycled crust, which had been altered by interaction with a CO2-, SO2-, and halogen-rich atmosphere.