Field experiments were conducted to test the seed yield responses of winter oilseed rape (Brassica napus L., cvs Libravo or Falcon) to the addition of different rates of S fertilizer, at three N application rates, on a sandy loam at Woburn, Bedfordshire, in 1990/91, 1991/92 and 1993/94. Large increases in seed yields, ranging from 0.7 to 1.6 t/ha, or 42-267% on a relative scale, were obtained in response to the application of 40 kg S/ha with 180 and 230 kg N/ha treatments. The effects of S were highly significant in 1991/92 (P < 0.01) and 1993/94 (P < 0.001) and close to significant (P = 0.053) in 1990/91. The yield benefits were obtained mainly from the application of the first 10 kg S/ha and further yield increases were unlikely above 40 kg S/ha. Increasing N application from 180 to 230 kg/ha decreased seed yield in 1990/91 and 1993/94, when no S was applied. In contrast, seed yield was not increased by S at zero or low (50 or 100 kg/ha) N rates. The interactions between N and S on seed yield were significant (P < 0.05) in 1990/91 but not in the other two seasons. Application of S also increased seed oil content in 1993/94, when the degree of S deficiency was particularly severe. With an application of 230 kg N/ha, the crops took up 5-22 kg S/ha at maturity when no S was applied and 26-51 kg S/ha when 40 kg S/ha was applied. The utilization efficiency of the fertilizer S ranged from 50 to 73% in the three seasons. Although the concentrations of total N in plants were largely unaffected by S treatments, large amounts of NO3-N accumulated in the leaves of S-deficient plants in 1993/94. This indicates that N metabolism was disrupted by S deficiency. The concentrations of S and the N:S ratios in different tissues and the whole plant changed considerably with time. The concentration of S in leaves at early flowering was found to be the best index in predicting S deficiency in terms of seed yield, and a critical value of 3.8 mg/g was obtained. In comparison, the N:S ratio in leaves at early flowering was a much poorer predictor of S deficiency.