The purpose of this study was to characterize high school chemistry students' ability to make translations between three representations of the structure of matter, and to determine the degree to which the students' ability to make these translations is related to reasoning ability, spatial reasoning ability, gender, and specific knowledge of the representations. Translation between formula, electron configuration, and ball-and-stick model representations of matter were chosen for study because of their promise for adding to knowledge of students' conceptual ecology, and because they may be of practical use for teaching and evaluation in chemistry classrooms. Representations have the characteristic that they embed selected details of the relevant concept or principle, but permit other details to fade. As one example, the chemical formula for water, H2O, explicitly conveys the identity of the constituent elements and their ratio, but does not explicitly convey the bond angle or whether the bonds are single or double. On the other hand, the ball-and-stick model of water explicitly conveys the bond angle and bond orders, but does not emphasize the ratio of the elements. Translation between representations is an information processing task, requiring understanding of the underlying concept to the extent that the individual can interpret the information provided by the initial representation and infer the details required to construct the target representation. In this study, the use of the translations of representations as an indicator of understanding of chemical concepts is developed in terms of (a) its relationship to four variables associated with achievement in chemistry, (b) specific representation error types, and (c) its utility in revealing details of students' conceptions and concept formation. Translation of representation performance was measured by administering, audio recording, transcribing, and scoring individual, task-based, think-aloud interviews. The associated interview schedule was entitled Translation of Representations—Structure of Matter [TORSOM]. Reasoning ability was measured by the Group Assessment of Logical Thinking—short form (GALT-s), spatial reasoning ability by the spatial reasoning subtest of the Differential Abilities Test (SRDAT), and prior knowledge of the representations by a test developed by the first researcher (Knowledge of Representations—Structure of Matter). When each of the hypothetical correlates were regressed on TORSOM individually, results indicated the KORSOM and GALT-s but not gender or SRDAT were statistically significant (alpha = .05). The two-predictor model accounts for 28% of the variance in the TORSOM scores. Representation error types are described and exemplified.