In France, about 95% of the fatty liver production comes from mule ducks, an infertile hybrid cross between female common ducks (Anas platyrhynchos) and Muscovy drakes (Cairina moschata). As a hybrid, genetic improvement of the performance of mule ducks is achieved by selection of the parental populations. Feed represents two-thirds of the total costs of duck production; thus, the breeders' main concern is to improve the birds' feed efficiency without impairing their capacity to be overfed after the growing period. To obtain insight into this economically important trait, we designed a divergent selection experiment for Muscovy sires on the basis of the residual feed intake (RFI) of their male mule progeny. Residual feed intake represents the fraction of feed intake that cannot be explained by maintenance and production requirements. Data comprised records from generations 0 to 2 with 227 Muscovy sires of 2,016 mule ducks. Growth and fat deposition were measured individually for all mule ducks. Half-sib mule ducks were placed together in 1 small pen, and the feed consumption was recorded for each pen; all half-sibs therefore had the same feed intake estimates. Heritability of RFI was high (0.83 ± 0.42), and after 2 generations of selection, the differences between lines in terms of RFI (contrast of 9.13 g/d, representing 0.76 phenotypic standard deviation; P < 0.0001) and feed conversion ratio (contrast of 0.08, representing 0.28 phenotypic standard deviation; P = 0.0003) were marked. In regard to carcass traits, the low-RFI line (high-efficiency animals) had heavier legs (476 vs. 463 g; P < 0.0001) and magret muscles (268 vs. 262 g; P = 0.001) than the high-RFI line (low-efficiency animals). No differences were observed for either the liver weight or the melting rate between the 2 lines (P > 0.46). Selection on RFI thus improves the feed efficiency of animals without impairing their capacity to produce fatty liver.