Structural equation modeling is an extension of multiple linear regression modeling, and primarily used to test the potential causality between variables. A model may involve unmeasurable latent variables (also called factors) which are hypothesized to represent the common concepts among their indicator variables (measured). Therefore, factor analysis and regression analysis are said to be accomplished in one step.;In this thesis, structural equation modeling is used in establishing regression models with latent variables to relate chemical structures to physical or chemical properties (e.g. boiling point), or chromatographic parameters. The chemical descriptors are usually correlated in a complicated manner, thus latent variables are employed to account for the intercorrelations among the measured variables. Also, structural equation modeling can test the correlation and/or causality among latent variables. Therefore, a plausible model can describe the factors which affect the observed chemical properties or phenomena, as well as the relationships among the factors.;Confirmatory factor analysis is a useful procedure that is subsumed under structural equation modeling. In this project, the underlying structure of a set of chemical descriptors is assessed by exploratory and confirmatory factor analysis.;In another study, structural equation modeling, as well as other two commonly known latent variable modeling techniques, Principal Component Regression (PCR) and Partial Least Squares (PLS), are applied to investigate acute lymphocytic leukemia with blood plasma analyzed by HPLC. The similarity and difference of these regression methods are described. |