In this article two kinds of error analyses are presented: (1) Error analysis using tests to describe errors and (2) cognitive error analysis in which one investigates the cognitive processes underlying the errors by thinking aloud. This method is carefully scrutinized because (1) it cannot reveal the causes of the errors, (2) it is inappropriate for testing hypotheses, (3) it does not lead to new hypotheses. Furthermore, its didactical value is disputable, because in most cases the relation between the teaching process and errors is not taken into account. Therefore, the epistemological value of cognitive error analysis must be questioned. For the future, a combination of developmental research and classical experimental research seems more promising than adhering to clinical approaches like cognitive error analysis. © 1986, GDM - Gesellschaft für Didaktik der Mathematik. All rights reserved.