To assess the quality of school education, much of educational research is concernedwith comparisons of test scores means or medians. In this paper, we shift this focus andexplore test scores data by addressing some often neglected questions. In the case ofBrazil, the mean of test scores in Math for students of the fourth grade has declinedapproximately 0,2 standard deviation in the late 1990s. But what about changes in thedistribution of scores? It is unclear whether the decline was caused by deterioration instudent performance in upper and/or lower tails of the distribution. To answer thisquestion, we propose the use of the relative distribution method developed by Handcockand Morris (1999). The advantage of this methodology is that it compares twodistributions of test scores data through a single distribution and synthesizes all thedifferences between them. Moreover, it is possible to decompose the total differencebetween two distributions in a level effect (changes in median) and shape effect(changes in shape of the distribution). We find that the decline of average-test scores ismainly caused by a worsening in the position of all students throughout the distributionof scores and is not only specific to any quantile of distribution.