Is it possible to predict relatively early in the life of children with disabilities those who are likely to develop peer problems if they get no extra help? This question is examined in our prospective study of a representative sample of children with hemiplegia attending mainstream schools. Hemiplegia is a particularly suitable model for studying integration as it is a relatively homogeneous condition that does not usually preclude mainstream placement. Our aim was to investigate whether the excess of peer problems could be predicted from information obtained some 4 years earlier. A representative sample of 55 children with hemiplegia in mainstream education was followed prospectively from a mean age of 7.1 years (Time 1) to a mean age of 10.7 years (Time 2). Using standardized measures of peer rejection, lack of friends, and victimization, two-thirds of the sample had at least one of these problems at Time 2. A greater number of peer problems was primarily predicted by two Time-1 variables: lower IQ and more teacher-reported externalizing problems (disruptiveness and hyperactivity). A risk index based on these two variables identified a high-risk subgroup that might particularly have benefited from early intervention to reduce behavioural problems, and nurture social skills. As more children with special needs are integrated into mainstream schools, it is increasingly important to remember that supporting these children requires appropriate provision to foster their social as well as their academic and physical development.