The results of more than 80 experiments on gastrointestinal parasitism and the impact of anthelmintic treatment on milk production in dairy cattle were reviewed. Abattoir surveys of culled dairy cows, faecal egg counts in milking cows, and serological tests and worm counts in cull cows in milk production studies were collated to assess the level of parasitism in dairy herds. The studies were divided into four general categories: induced infections in previously uninfected cattle; naturally infected cattle treated in mid-lactation; naturally infected cattle treated one to three times during the dry period and/or just before or just after parturition; and naturally infected cattle treated repeatedly from early lactation or given strategic treatments throughout the year. In most studies, the milk production of anthelmintic-treated cattle was compared with that of untreated controls. The anthelmintics investigated included members of the organophosphate, benzimidazole, imidazothiazole and macrocyclic lactone groups. The number of experiments in which the medicated (or uninfected) group had a higher milk yield was compared with the number of experiments in which the control (or infected) group had a higher yield. Overall, the studies demonstrated that grazing dairy cattle are likely to be infected with gastrointestinal nematode parasites, usually Ostertagia ostertagi and Cooperia species. These infections may be present as inhibited larvae, and a periparturient or spring rise is associated with their emergence. There is, at present, no reliable means of determining whether a cow or a herd may be parasitised subclinically at a level sufficient to interfere with milk production. In 70 of 87 experiments (80 per cent) there was an increase in milk production (P < 0.001) after anthelmintic treatment, with a median increase of 0.63 kg/day. In each of the four trial categories, a majority of the studies showed that anthelmintic treatment increased milk production. The yield of milk fat by the medicated cows was greater than by the controls in 26 of the 35 experiments in which that variable was studied (P < 0.01).