The analysis and synthesis of plucked-guitar tones via source-filter approximations is a popular and established method for modeling the resonant behavior of the string as well as the driving excitation signal. By varying the source signal, a nearly unlimited number of unique tones can be produced using a given filter model. However, it as unclear as to how exactly the model excitation signals should be parameterized in order to capture the nuances of a guitarist's articulation from a recorded performance. In this paper, we apply principal components analysis to a corpus of excitation signals derived from plucked-guitar recordings in order to design a codebook that captures the unique characteristics of certain string articulations. The development of an excitation codebook has several applications, including expressive synthesis of guitar tones for virtual music interfaces and insight into the expressive intentions of a performer through audio analysis.