OBJECTIVE: This study is to identify predictors of loss to follow-up among adults with head injury. STUDY DESIGN AND SETTING: A prospective cohort of 1,857 adults enrolled in the Medical Research Council (MRC) CRASH trial known to be alive 2 weeks after head injury. Six-month follow-up was defined as "overdue" if over 6 months late. Patient information collected at enrollment and after 14 days was used to predict overdue follow-up. A random two-thirds of the cohort was analyzed using logistic regression and binary recursive partitioning. The regression model and decision rule derived by recursive partitioning were evaluated using the remaining third. RESULTS: Overdue follow-up was more likely in patients aged 25-34 years (odds ratio, 1.76; 95% confidence interval [CI]=1.18-2.62), victims of assault (1.63; 1.09-2.45), patients independent after 2 weeks (1.79; 1.18-2.72) and patients for whom postcodes (2.36; 1.65-3.39), telephone numbers (1.82; 1.19-2.79) or general practitioners (1.67; 1.16-2.39) were unknown. Binary recursive partitioning specifically identified males aged younger than 43 years to be at risk. CONCLUSION: Successful follow-up in head-injury studies requires patients' postcodes and telephone numbers to be available. Young men remain at risk of becoming lost to follow-up, presenting a challenge for researchers aiming for complete data.