We aimed to discover, through a controlled experiment, whether cognitive and non-cognitive assessment would select higher-achieving applicants to medical school than selection by lottery. We carried out a prospective cohort study to compare 389 medical students who had been admitted by selection and 938 students who had been admitted by weighted lottery, between 2001 and 2004. Main outcome measures were dropout rates, study rate (credits per year) and mean grade per first examination attempt per year. Study rates in the 4 pre-clinical years of medical school were used to categorise students' performance as average or optimal. Pre-admission variables did not differ between the two groups. The main outcome of the selection experiment was that relative risk for dropping out of medical school was 2.6 times lower for selected students than for lottery-admitted controls (95% confidence interval 1.59-4.17). Significant differences between the groups in the percentage of optimally performing students and grade point average for first examination attempts were found only in the 2001 cohort, when results favoured the selected group. The results of the selection process took into account both the assessment procedure involved and the number of students who withdrew voluntarily. This is the first controlled study to show that assessing applicants' non-cognitive and cognitive abilities makes it possible to select students whose dropout rate will be lower than that of students admitted by lottery. The dropout rate in our overall cohort was 2.6 times lower in the selected group.