The electrochemical conversion of CO2 and H2O into liquid fuel is ideal for high-density renewable energy storage and could provide an incentive for CO2 capture. However, efficient electrocatalysts for reducing CO2 and its derivatives into a desirable fuel are not available at present. Although many catalysts can reduce CO2 to carbon monoxide (CO), liquid fuel synthesis requires that CO is reduced further, using H2O as a H+ source. Copper (Cu) is the only known material with an appreciable CO electroreduction activity, but in bulk form its efficiency and selectivity for liquid fuel are far too low for practical use. In particular, H2O reduction to H2 outcompetes CO reduction on Cu electrodes unless extreme overpotentials are applied, at which point gaseous hydrocarbons are the major CO reduction products. Here we show that nanocrystalline Cu prepared from Cu2O (‘oxide-derived Cu’) produces multi-carbon oxygenates (ethanol, acetate and n-propanol) with up to 57% Faraday efficiency at modest potentials (–0.25 volts to –0.5 volts versus the reversible hydrogen electrode) in CO-saturated alkaline H2O. By comparison, when prepared by traditional vapour condensation, Cu nanoparticles with an average crystallite size similar to that of oxide-derived copper produce nearly exclusive H2 (96% Faraday efficiency) under identical conditions. Our results demonstrate the ability to change the intrinsic catalytic properties of Cu for this notoriously difficult reaction by growing interconnected nanocrystallites from the constrained environment of an oxide lattice. The selectivity for oxygenates, with ethanol as the major product, demonstrates the feasibility of a two-step conversion of CO2 to liquid fuel that could be powered by renewable electricity.