In this work, modeling and monitoring of end-use power consumption in commercial buildings are investigated through both Top-Down and Bottom-Up approaches. In the Top-Down approach, an adaptive support vector regression (ASVR) model is developed to accommodate the nonlinearity and nonstationarity of the macro-level time series, thus providing a framework for the modeling and diagnosis of end-use power consumption. In the Bottom-Up approach, an appliance-data-driven stochastic model is built to predict each end-use sector of a commercial building. Power disaggregation is studied as a technique to facilitate Bottom-Up prediction. In Bottom-Up monitoring and diagnostic detection, a new dimensionality reduction technique is explored to facilitate the analysis of multivariate binary behavioral signals in building end-uses.