The study investigated the extensive writing performance of the students across three proficiency levels of English: Advanced, Intermediate and Elementary. From each level, three students were selected purposively to represent high-, middle- and low-achievers. Their nine pieces of writing were then assessed by means of the theory of patterns of cohesion (Halliday and Matthiessen, 2004; Eggins, 2004; Paltridge, 2006; Salkie, 1995; and Gerot & Wignell, 1994). The patterns consist of five cohesive devices: reference, conjunction, lexical cohesion, ellipsis and substitution. The assessment focused on analyzing the students’ failures in observing the patterns. The study, therefore, resembles an error analysis but adopts the cohesive devices, instead of linguistic/ grammatical classifications or other kinds of taxonomy as the sources of errors. As such, the study is the first that introduced the uses the devices as the sources of errors. Further, the errors made by the learners were calculated and compared with the total number of the words they wrote in their essay writing. This way, the data concerning the frequency of the failures in the cohesive devices across the students’ levels were obtained, tabulated and then discussed. Based on the results, a model of an analytic scoring guide was proposed to be used for assessing the students’ writing performance by considering the differences of levels. The model is expected to be useful for writing teachers as it has a higher level of practicality and measurability compared with other available analytic scoring methods.

Key words: extensive writing, analytic scoring guide, cohesion, cohesive devices, error analysis, taxonomy