The primary purpose of this study is to investigate how well vocabulary features, such as collocation, word family, etc., predict vocabulary scores in writing graded by human raters, and the degree to which they explain holistic scores. Forty-nine writings by first-year non-native English- speaking international students at a mid-western U.S. university were analyzed by using two multiple regression models. Data includes both holistic and analytic vocabulary scores given by four human raters and specific vocabulary features counted by concordance programs. Holistic and analytic vocabulary scores were estimated using many-facet Rasch measurement analysis. It was found that (a) collocation variety and essay length (or tokens) explained 32.0% of variance in the holistic writing scores, and (b) collocation variety and word family accounted for 29.5% of the variance in the vocabulary scores. The results of this study suggest that it is beneficial to teach and learn vocabulary knowledge to improve students’ writing proficiency.