Before 1970, SLA research was dominated by Contrastive Analysis research. The purpose of this research was to test the Contrastive Analysis Hypothesis’ idea that learners’ errors could be predicted on the basis of differences between the learners’ first language and the target language. Linguists compared languages to see what the differences were and then used those data to predict the transfer errors learners would make. It was hoped that the data from linguists could eventually be used to help design special drills and exercises that would help learners learn accurate use of second languages but not make any errors while doing so. The prevailing view of errors at that time, was that errors were bad habits that must be broken and not reinforced. They were viewed as harmful. Experts felt that errors had to be avoided, even in the course of learning, and for decades SLA researchers worked toward this goal.

By 1970, it was clear that Contrastive Analysis could not predict the errors learners would make; therefore, researchers concluded that their must be some other processes involved in second language learning besides interference. As researchers discovered many errors that were clearly not due to interference, it became safe to assume that there must be other sources of errors besides the first language. Researchers then, shifted their focus from predicting errors based on contrasting languages, to classifying the various kinds of errors they saw learners making. It was hoped that by studying the various types of errors that learners made at various stages of learning, that researchers could get a clearer view of the second language learning process. Thus, Error Analysis, the study of learner language for the purpose of classifying errors and identifying their sources, emerged as the dominant SLA research.

Whereas Contrastive Analysis was based only on the assumption that errors were all due to first language interference and were somehow harmful to the learner’s development, Error Analysis was based on the assumption that errors were a natural and healthy part of the language learning process – a natural by-product of the learners step-by-step discovery of the second language’s rules through a process of trail and error. This process was called "Creative Construction".

During creative construction, learners look to different sources for information about the target language. Sometimes the learner simply assumes that features of her first language are the same as those of her target language, and so, transfer might occur as she tests out her assumptions by using L1 features in the L2. This could be either successful or unsuccessful. In other times, a learner might vaguely know a rule of the target language and attempt to use it creatively. She might for example, generalize the rule to a new context she is unsure of. At times such creativity might be successful, at other times it may lead to an error. Learners could also get ideas from textbooks, movies, dictionaries, and other such materials and then test those ideas out when speaking or writing, sometimes resulting in errors, and sometimes resulting in accurate or successful use of the target language. In sum, learners could get ideas about the target language from many different sources, including themselves, and these ideas could result in both progress and errors. In that sense, the sources of errors are also the sources of the learners’ knowledge.

Errors viewed in that light, are now seen as a window to the learning process for researchers. Errors are no longer viewed as mindless behavior such as "bad habits", but actually considered to be the by-product of the learner’s intelligent efforts to discover the rules of the target language. Simply put, learners, with their incomplete and inaccurate knowledge about the target language, can reasonably be expected to produce incomplete and inaccurate utterances. Errors are in a sense, a type of knowledge the learner has about the target language. They are the hypotheses she is currently experimenting with. For researchers, learner errors can tell them about the learner’s internal system of knowledge and the rules she follows. In fact, errors are often more useful than accurate language use because accurate language use could come from anywhere - it is impossible to guess the source of accurate language use. An accurate sentence produced by the learner could simply have been memorized, but an inaccurate one is probably the result of misapplying some rule. Errors often have a certain character that gives away their source. For example, it is possible to guess whether an error has come from the first language if it seems similar to the first language. Likewise, errors that do not resemble the first language but resemble those made by children can be attributed to the natural process of development. An utterance with a missing piece tells the researcher that the learner’s system is still missing some pieces.