Nitrogen (N) inputs and outputs were measured over 3 years in a trial with four farmlets (each with 16 randomly-allocated 0·4 ha paddocks) on permanent white clover/ryegrass pastures which were grazed throughout the year by dairy cows near Hamilton, New Zealand. Three farmlets were stocked at 3·3 cows/ha and received nominal rates of N fertilizer (urea in 8–10 split applications) of 0, 200 or 400 kg N/ha per year. A fourth farmlet with 4·4 cows/ha received 400 kg N/ha per year and was supplemented with maize grain during the first two years.Nitrogen balances were calculated, with [sum ]N inputs[ape ][sum ]N outputs. Annual inputs from N2 fixation were 99–231 kg N/ha in the 0 N farmlet, but declined to 15–44 kg N/ha in the 400 N farmlets. The main N outputs (in kg N/ha per year) were in milk (72–126), nitrate leaching (20–204), and transfer of N via cow excreta from pastures to lanes and milking shed (54–92). Gaseous losses by denitrification (3–34) and volatilization (15–78) were smaller than the other N outputs but increased significantly with N fertilizer application. In the maize-supplemented farmlet, N outputs in milk were 31% higher than in the corresponding non-supplemented 400 N farmlet, whereas leaching losses averaged 17% lower during the 2 years of supplementation.In the N-fertilized farmlets, estimated N balances were influenced by inclusion of the transitional N processes of immobilization of fertilizer N into the soil organic N pool (estimated using 15N at 42–94 kg N/ha per year) and the contribution from mineralization of residual clover-fixed N in soil not accounted for in the current estimates of N2 fixation (estimated at up to 70% of measured N2 fixation or 46 kg N/ha per year). However, these processes were counteracting and together were calculated to have only a small net effect on total N balances.The output of N in products (milk, meat and feed) relative to the total N input averaged 26% in the 400 N farmlets, and is compared to that measured for commercial intensively-managed dairy farms in England and the Netherlands (14–20%). The 0 N farmlet, which was reliant on N2 fixation as the sole N input, was relatively very N-efficient with the milk production being 83% of that in the 400 N farmlet (at 3·3 cows/ha) and the N output in products relative to total N input averaging 52%.