Monitoring educational changes over many years is problematic when there are differences in curricula, the nature of the variables being measured, and the selection of participants. Rasch measurement techniques provide a procedure that enables each of these issues to be examined. Using archived and specially collected data, tests of numeracy undertaken in Tasmania over a 20-year period, from 1978 to 1997, were equated and mapped onto the same continuum through a combination of common item and common person equating. Examination of fit to the model showed that the nature of the measured construct had not changed over this period. Although test difficulty appears to have risen over the period, student achievement was relatively unchanged. The implications of these findings for longitudinal studies of achievement are discussed.