Improving feed efficiency (FE) is key to reducing production costs in aquaculture and to achieving sustainability for the aquaculture industry. Feed costs account for 30–70% of total production costs in aquaculture; much work has been done on nutritional and husbandry approaches to improve FE but only a limited amount of research has been devoted to using genetics, despite its potential. This paper reviews past work to improve FE in fish using selective breeding and assess future directions. Direct selection on FE traits requires methods to measure individual feed consumption and estimate FE efficiently and accurately. This is particularly difficult to do in fish because of the environment in which they live. Many of the published studies on FE were found to be inaccurate because of methodological problems. The relatively low heritability estimates of FE traits in fish published to date are probably partly as a result of inaccurate measurements of feed intake. Improving ways to measure the individual feed intake with high accuracy will be critical to the successful application of genetics to improving FE. Indirect selection criteria that could be used to improve FE (including growth after starvation/refeeding, body composition, neuropeptides or hormone levels) are discussed. Promising approaches to measuring feed intake accurately that may enable these studies to be undertaken are identified. More work using these will be needed prior to assessing the practicality of the introduction of direct or indirect traits for FE in fish genetic improvement programmes.