We reviewed all randomised trials on cognitive rehabilitation in order to determine the effective elements in terms of patients' and treatment characteristics, treatment goals and outcome. A total of 95 random controlled trials were included from January 1980 until August 2010 studying 4068 patients in total. Most studies had been conducted on language (n = 25), visuospatial functioning (n = 24), and memory (n = 14). Stroke patients were the commonest subjects (57%; overall mean age = 52.2, SD = 15.0 years). Of the interventions 39% were offered more than 12 months after onset and 23% were offered within two months of onset. The mean (SD) number of hours of treatment actually delivered was 4.1 (3.6) per week; treatment was mostly offered individually. No papers gave specific information on the expertise or competences of the staff involved. With 95 RCTs there is a large body of evidence to support the efficacy of cognitive rehabilitation, and the current study can serve as a database for clinicians and researchers. But most studies have given little information about the actual content of the treatment which makes it difficult to use the studies when making treatment decisions in daily clinical practice. We suggest developing an international checklist to make standardised description of non-pharmacological complex interventions possible.