This review charts the use of the concept of harvest index in crop improvement and physiology, concentrating on the literature from the last 20 years. Evidence from abstract journals indicates that the term has been applied most to small grain cereal crops and pulses, in India, Western Europe and the USA, and that it has been less useful for maize and tuber crops. Standard methods of measuring harvest index, the associated problems of measurement and interpretation, and representative values for a range of world species are reviewed. The values for modern varieties of most intensively-cultivated grain crops fall within the range 0.4 to 0.6. Variation between varieties of the same species is illustrated by trends in the harvest indices of old, outclassed and recent varieties of temperate and mediterranean wheat and barley (compared under uniform conditions); this shows a progressive increase throughout the present century, although improvement has been much slower in Australia and Canada than in the UK. In most cases, the improvement in harvest index has been a consequence of increased grain population density coupled with stable individual grain weight. The high heritability of harvest index is explored by examining its (rather weak) response to variation in environmental factors (fertilisation, population density, application of growth regulators) in the absence of severe stress. A fuller perspective is gained by reviewing aspects of the harvest index of rice, maize and tropical pulses. With rice, attention must be paid to the fact that the adhering lemma and palea (not primarily part of economic yield) can make up 20% of grain weight; and there are important interactions among biomass, grain yield and season length. Maize differs from most small grain crops in that harvest index (in N. American varieties) was already high at the start of this century, and increases in yield potential have been largely the consequence of increased biomass production. The harvest index of many pulse species and varieties tends to be low because selection has been for some yield in all seasons. Extension of the harvest index concept to express the partitioning of mineral nutrients as well as dry matter (e.g. the nitrogen harvest index) has provided a range of responses whose implications for production and breeding remain to be explored. It is concluded that even though the principal cereal crops appear to be approaching the upper limit of harvest index, and future yield gains will have to be sought by increased biomass production, there will still be a need for the concept of harvest index as a tool in interpreting crop response to different environments and climatic change.