In this study, structural equation modeling (SEM) is applied to an instrument assessing students' understanding of the particulate nature of matter, the collective properties and physical changes, such as melting, evaporation, boiling and condensation. The structural relationships among particular groups of items were investigated. In addition, three cognitive variables, such as logical thinking, field-dependence/field-independence and convergence/divergence dimension were included in the SEM analysis and their effects on students' performance were estimated. Specifically, three models were tested: a confirmatory factor model (CFM), a multiple-indicator multiple-cause (MIMIC) model and a path analysis. The results showed that the three cognitive variables, along with achievements in the dimensions of structure understanding, sufficiently explain students' understanding of physical changes, providing additionally their direct and the indirect effects. Moreover, a theoretical analysis and interpretation of the results are provided that adds to our understanding about the role of cognitive variables in the mental processes involved in learning the specific-domain material. Implications for science education are discussed. (Contains 2 tables and 4 figures.)