The examination of the official records from Artas, Isère (North Dauphiné, France), provides nearly continuous information from 1540 to 1900. These allowed us to study the twinning during this period, to specify its importance, its evolution, and its links with diverse other characteristics of reproduction. The number of twins per 1000 births is on a general average very close to 11. It seems to decrease during the last four centuries. However, the variability is very high from one year to another or one decade to another, including the composition of twins (identical or different sex), with the result that no statistical argument allows us to reject the null hypothesis. Numerous data have been collected that support a partially genetic determinism in twinning. The number of children in families with twins is clearly higher that in other families. Age of death of the mother of twins is higher than that of the mother of single births. With the result that 'gemellity', fecundity and longevity are related. On the other hand, the perinatal and infant death rate is much higher for twins than for single births, so that the percentage of 'useful' children is very low for twins. There is therefore, at the same time, selection and counter-selection. The momentary importance of either could explain the extent of variation of the 'gemellity' rate. The compensations, at short or long time, between the two processes could explain the roughly constant rate of 'gemellity' during the last four centuries.