This paper proposes a broader and more rigorous approach to the analysis of errors produced by second language learners. As a supplement to longitudinal studies, a procedure called "lateralisation" is advocated as a way of providing the researcher with more data than is normally available through examination of learner text alone. In this procedure a preliminary analysis of learner texts is used to draw preliminary hypotheses, which can then be tested by giving the learner reconstructed errors for his approval or rejection. From this, an abstract characterization of errors can be made. The elicitation of linguistic and metalinguistic statements from the learner on his errors is the stage that should considerably increase the amount of specific data the researcher can obtain. However, since the learner's language is in a state of flux, it is important to use the elicitation techniques very soon after the initial learner texts have been developed. Although this work is limited to linguistic analysis, the help of social psychology is also indicated as important. Several examples of elicitation are given, including a case study that illustrates the use of elicitation in error analysis. (AM)