Second Language Acquisition (SLA) researchers have yet to map the developmental stages language learners go through as they approach the target language. In studies of ESL writing, the term "advanced learner" has been applied indiscriminately to learners ranging from freshman ESL composition to graduate students (Bardovi-Harlig and Bofman, 1989; Chaudron and Parker, 1990; Connor and Mayberry, 1996; Hinkel, 1997, 2003). There is a need to examine the advanced stages of SLA in order to refine SLA theories and pedagogical approaches.A corpus of texts written by eleven graduate students in applied linguistics who are non-native-English speakers from several linguistic backgrounds was analyzed to determine the texts' lexical, morphological, and syntactic fluency, accuracy, and complexity. A sub-corpus of papers by seven native-English-speaking peers was used for comparison. The texts were sit-down and take-home examinations written in a doctoral program at the end of the first semester and three years later. Surveys and interviews were conducted to supplement the corpus with ethnographic data.This dissertation defines data-based criteria that distinguish four quantitatively and qualitatively distinct developmental stages: the advanced, highly advanced, near-native, and native-like stages. Advanced learners make more frequent and varied errors (with articles, prepositions, plural and possessive markers, agreement and anaphors), which can be explained by linguistic transfer. Native-like writers make few errors that can be explained by overgeneralization of conventions from informal English and working memory limitations (just like native speakers' errors). Throughout the four stages, errors (i.e., incorrect forms that reflect lack of linguistic knowledge (Corder, 1967)) became less frequent, and more of the incorrect usages appeared to be mistakes (occasional slips).This dissertation supports Herschensohn's (1999) proposal that SLA is a process of transfer followed by relearning of morpho-syntactic specifications. Syntax was used with the greatest accuracy (Bardovi-Harlig and Bofman, 1989), while lexicon (especially function words) was the weakest. In addition, length of stay in an English-speaking country and amount of interaction with native speakers were proportional with accuracy. An important pedagogical recommendation is that (corpus-assisted) language teaching should continue until the target language is reached.