With the increasing use of integrated tasks in assessing writing, more and more research studies have been conducted to examine the construct validity of such tasks. Previous studies have largely focused on reading-writing tasks, while relatively little is known about graph-writing tasks. This study examines second language (L2) writers' test-taking strategies in relation to their performance on a graph-writing test administered to English learners in health science and medical majors. Data from a strategy inventory and open-ended questions were collected and analyzed to determine if the task elicited academic writing strategies and to identify construct-related issues. The results from structural equation modeling analyses indicated that writers were engaged in graph comprehension, graph interpretation, and graph translation strategies during the task, and each type of strategy use generally had a positive impact on their test performance. The qualitative results pointed to L2 writers' difficulties in lexical knowledge and use, as well as possible sources of construct-irrelevant variance including graph familiarity, topical knowledge, and test-wiseness strategy use. The findings have theoretical and practical implications for the development and use of graph-writing tasks. (Contains 5 tables and 2 figures.)