Mice were selected for efficiency of feed utilization (gain/food) over two age ranges from 3 to 5 weeks and from 5 to 7 weeks. Selection was conducted under ad libitum feeding and, separately, under a regime where each animal was given the same amount of food each generation, which was the mean food consumption of the bone population fed ad libitum. Selection was within families, each selection line was replicated and two unselected controls were also kept. After either 7 or 8 generations of selection, each line was tested on the alternative feeding regime. Carcass analyser of animals fed ad libitum from all lines were carried out in the succeeding generation. There was a response to Feleotion in feed efficiency under both feeding regimes, in both age intervals. The realized heritabilities were 0.12 and 0.17 in the 3-5 week lines, and 0.11 and 0.12 in the 5-7 week lines for the fixed and ad libitum lines respectively. These estimates are the average for the two replicates in each treatment. The ad libitum lines were more efficient than the fixed lines, and within each line, males were more efficient than females. Improvements in feed efficiency were attributed entirely to the increased rate of gain, there being no change in food consumption. Weight at start of test tended to be reduced in the selected 3-5 week lines, but not consistently. No interaction for feed efficiency was found between the feeding regime used for selection and that used for subsequent testing. Selection for feed efficiency caused an increase in percent ether extract over the levels in the controls. Corresponding to this increase, there was a reduction in percent water. Various aspects of feed efficiency and its relationship with other traits are discussed. It was concluded that feed efficiency and rate of gain are highly and positively correlated, both phenotypically and genetically. In view of the slow response, it was suggested that selection for feed efficiency may not be worthwhile because of the extra labour and facilities required, as similar gains may be obtainable by selection for gain alone.