"The invention of sophisticated measurement tools, such as motion capture devices, handheld computers, Bluetooth devices, eye-trackers, and brain scanners, has facilitated the collection of functional data that can be considered to arise from an underlying smooth function varying over a continuum such as time and space. Functional data analysis (FDA) is an emerging branch of statistics, which develops and applies statistical methods for the analysis of such types of data. Various FDA methods have been proposed by extending traditional multivariate statistical methods to accommodate functional data. Nonetheless, there has been little attempt to develop functional extensions of structural equation modeling (SEM), in spite of the remarkable popularity of SEM in various disciplines due to its flexibility of modeling complex relationships among observed and latent variables. This thesis thus aims to propose a general framework for functional SEM, called functional generalized structured component analysis (functional GSCA), to examine a variety of hypothesized relationships among observed and latent variables, while permitting observed variables to be functional rather than multivariate. The thesis begins by describing GSCA and penalized least squares smoothing as the two basic building blocks of the proposed method. Subsequently, it provides the technical details of the proposed method. The model for functional GSCA is provided and a penalized least squares criterion is developed for parameter estimation, which is minimized by an alternating penalized least squares algorithm. The thesis also demonstrates the usefulness of the proposed method by analyzing synthetic and real data sets. It concludes with discussions on limitations and possible extensions of the proposed method." --