Few, if any, of children's behavioural or cognitive characteristics assessed in the first years of life demonstrate stability until later childhood; early characteristics have so far failed to show an association with future psychopathology. This longitudinal study, from 4–8 months to 4.7 years old, focused on stability and change of extreme temperamental traits in groups of infants subselected from a large birth cohort. Persistent extreme temperament at four and eight months old did not increase stability of temperament to four years of age, relative to other children in the whole population. Sizeable change occurred, and the environmental parameters associated with negative temperamental change did not seem to be the same as those related to positive change. Boys with extreme scores were more stable, while girls appeared more prone to positive change. It is hypothesised that the direction of temperamental change in the first years could be more meaningful for long-term prediction of disorders than any one assessment of temperament taken at any one year.