in this talk, i focus on factor analysis of high-dimensional time series data, in which the dimension of data is allowed to be even larger than the length of data. analysis of high-dimensional data suffers from the curse of dimensionality. factor analysis is considered as an effective way for dimension reduction. factor models presume that a few common factors can explain most of the variation/dynamics of an observed process in high dimensions. in the models, factor loadings are introduced to reflect the percentages of variations explained and contributions made by these common factors. based on real data analysis, it has been discovered that the loadings may vary in different situations/regimes. to interpret this observation and capture the regime-switching mechanism often encountered in practice, we propose a threshold factor model for high-dimensional time series data, in which a threshold variable is introduced to distinguish different regimes. loadings controlled by the threshold variable vary across regimes. the theoretical properties of the procedure are investigated.