Some children with juvenile idiopathic arthritis either do not respond, or are intolerant to, treatment with disease-modifying antirheumatic drugs, including anti-tumour necrosis factor (TNF) drugs. We aimed to assess the safety and efficacy of abatacept, a selective T-cell costimulation modulator, in children with juvenile idiopathic arthritis who had failed previous treatments. We did a double-blind, randomised controlled withdrawal trial between February, 2004, and June, 2006. We enrolled 190 patients aged 6-17 years, from 45 centres, who had a history of active juvenile idiopathic arthritis; at least five active joints; and an inadequate response to, or intolerance to, at least one disease-modifying antirheumatic drug. All 190 patients were given 10 mg/kg of abatacept intravenously in the open-label period of 4 months. Of the 170 patients who completed this lead-in course, 47 did not respond to the treatment according to predefined American College of Rheumatology (ACR) paediatric criteria and were excluded. Of the patients who did respond to abatacept, 60 were randomly assigned to receive 10 mg/kg of abatacept at 28-day intervals for 6 months, or until a flare of the arthritis, and 62 were randomly assigned to receive placebo at the same dose and timing. The primary endpoint was time to flare of arthritis. Flare was defined as worsening of 30% or more in at least three of six core variables, with at least 30% improvement in no more than one variable. We analysed all patients who were treated as per protocol. This trial is registered, number NCT00095173. Flares of arthritis occurred in 33 of 62 (53%) patients who were given placebo and 12 of 60 (20%) abatacept patients during the double-blind treatment (p=0.0003). Median time to flare of arthritis was 6 months for patients given placebo (insufficient events to calculate IQR); insufficient events had occurred in the abatacept group for median time to flare to be assessed (p=0.0002). The risk of flare in patients who continued abatacept was less than a third of that for controls during that double-blind period (hazard ratio 0.31, 95% CI 0.16-0.95). During the double-blind period, the frequency of adverse events did not differ in the two treatment groups. Adverse events were recorded in 37 abatacept recipients (62%) and 34 (55%) placebo recipients (p=0.47); only two serious adverse events were reported, both in controls (p=0.50). Selective modulation of T-cell costimulation with abatacept is a rational alternative treatment for children with juvenile idiopathic arthritis. Bristol-Myers Squibb.