Migration of Gammarus pulex pulex (Linnaeus, 1758), G. fossarum Koch in Panzer, 1836, and Echinogammarus berilloni (Catta, 1878) has been studied in a small French chalk stream, the Slack. Three different approaches to investigate both up- and downstream migration were used: (1) migration survey, with a sampling program of migration at intervals of two weeks or a month at twelve localities in the river Slack; (2) continuous measurement of migration at three habitats with very stable, normal and very unstable environmental conditions, respectively, lying within 100 m of one another and populated by the same species, G. fossarum; (3) finally, marking experiments in order to identify and trace animals with a given behaviour. Both drift and upstream migration show a considerable microgeographic variation, which is larger for Gammarus than for E. berilloni. During the relatively warm year of 1975, the migration activity of E. berilloni was stronger than in 1974. Upstream migration was concentrated in early summer, while drift fluctuated during the year. Most animals migrated during the night, although the diel variation in drift was quite different from that in upstream migration. Water temperature and its diel fluctuations have a large effect on non-accidental migration. Changes in chemical composition of the water seem to be important as well. Light conditions have only a slight influence on migration patterns. Physical disturbance of the riverbed (for instance by wading cows or the scouring effect of spates) influences migration rather negatively. The mean size of migrating animals was larger than the average size of the standing crop. Upstream migrants were larger during hours of high upstream migration activity, while the animals that drifted in peak hours were usually smaller than those drifting in hours of low activity. Both up- and downstream migration proved to be a constant behaviour; most drifters of a particular night drifted again the following night and most upstream migrants moved again upstream after they had been marked. In particular our results on microgeographic and seasonal variation show clearly that a quantitative approach to migration would have been premature. Secondly, they make a direct correlation between production and drift unrewarding. The continuous measurement of migration showed that for this type of investigation field work is preferable to laboratory experimentation, since it gives more reliable results than those achieved under laboratory conditions.