Model reduction of large scale power systems is currently carried out by coherency-based methods which start from a large data set of network model parameters. This monograph summarizes the most basic notions that will enable model reduction of large power systems using only phasor measurement unit (PMU) data. The monograph starts by reviewing the general aspects of network model-based reduction methods and describing the corresponding network aggregation techniques. For the measurement-based model reduction methods, observations are made about the voltage and frequency oscillation amplitude from the physical characteristics and from the time-domain simulations. These observations are illustrated with time-domain simulations of radial power systems with different network, machine, and exciter models. Reduced-order models of radial systems are obtained with simplified algorithms and are compared to the ones obtained by aggregation techniques. A numerical algorithm for parameter extrapolation is also outlined.