The present study is inspired by the often heard Chinese university level students’ complaint that they do not improve in English proficiency during their university courses. With a pre-post design, the study explores the potential gains in language development in free response data (writing samples) of three groups of L2 learners: a senior high school group and two university groups of different proficiency levels. Four writing samples, two collected at the beginning and two at the end of the students’ respective courses, were scored holistically on general proficiency and analytically on 47 complexity measures in a computerized tool (Synlex Analyzer). The holistic scores showed some improvement over time for the high school group, but not for the university groups. The analytical measures showed improvements in fairly different aspects of the written language for the three groups, suggesting that at different levels of proficiency different variables may develop. The highest level group actually regressed in almost all syntactic variables, but additional hand coded measures point to a subtle move toward a more academic style of writing with more non-finite constructions. The findings suggest that no single complexity measure is robust for all proficiency levels and that for the highest levels, other metrics tapping into inter-clausal complexity should be added.