Based on the framework of academic writers’ evaluator identity construction via citation, this paper selects 40 international research articles of both natural and social sciences, and compares the evaluator identity via citation from two perspectives: the dimension of evaluation and selection of lexical devices. Overall, the results show that evaluative citations occur frequently in texts of both natural and social sciences, most of which use the non-intergral structure to convey a positive attitude towards the results of previous studies in an impersonal and objective manner. The results also demonstrate that research articles of social sciences generally use more evaluative citations than those in natural sciences. The reason maybe that natural sciences are featured by objectivity and impartiality and focus more on originality of methods and results whereas social sciences are featured by interdisciplinarity and complexity of readers’ background and highlight importance of previous studies (especially those conducted by authority figures) to demonstrate the research value. Meanwhile, social science writers tend to establish a more negotiable interpersonal relationship to gain peer identification.