Error analysis (EA) is the systematic investigation of the types and causes of errors made by speakers and writers in their non-native languages (L2). Interest in EA as a systematic branch of study arose in the late 1960s, in close connection with contrastive analysis (CA), which had developed a little earlier. EA has been used for two different purposes, pedagogical and psycholinguistic. The pedagogical aim is, as it was for early CA, to provide feedback to the teacher about material and methods. More important today, however, is the psycholinguistic aim of EA: to illuminate how languages are learnt and produced. EA may provide a window for observing what goes on in the learner’s mind. EA has three different stages: identification, classification and explanation of the errors in a corpus. EA has been frequently used in the areas of phonology, morphology, syntax and lexis. Discourse analysis, on the other hand, has made little direct use of the concept of error. The term ‘error’ has a negative ring. It implies something undesirable and avoidable in the learner’s language. However, EA has, perhaps somewhat paradoxically, contributed a great deal to creating a different and sounder perspective: errors are, in fact, normal and inevitable features of the process of language learning. The study of learner errors should primarily be taken as evidence not of failure to conform to L2-usage but of success and achievement in the course of the learning process. It is obvious that EA has limitations. Since it is more concerned with what learners cannot do than with what they can do, it can be said to represent an incomplete and one-sided approach to language learning. Also, it cannot cope very well with the problem of avoidance: how and why learners avoid particular words, phrases and constructions. Its immediate chronological successor, interlanguage studies, which has largely merged with second language acquisition studies in general, was a natural consequence of the insufficiencies of EA. All the same, EA, which had its heyday in the 1970s but experienced a decline in the 1980s and early 1990s, has been of importance in the development of second language acquisition research. The main criticism, its insufficiency, can in fact be leveled at most studies of SLA. If we are aware of its limitations, the study of errors, their causes and development can provide valuable information on the underlying learning processes. Future access to large computerized corpora, especially the ICLE corpus (International Corpus of (Advanced) Learner English) which is now being collected (see Granger 1994), will no doubt give a new impetus to EA, since the analysis can be automatic or semi-automatic. The ICLE corpus will, when it is completed, provide an invaluable source for investigations of both errors and advanced learner language in general.