A general technique is proposed for estimating rupture geometry from measurements of body wave durations, wherein mesurements of the overall signal duration or the duration of a single pulse identified in all of the arrivals are inverted to estimate the geometry of the overall rupture or of a specific subevent, respectively. The complexity of a waveform is determined by comparing the overall signal duration to the duration of the most prominent pulse in the waveform; this complexity is used to estimate the ratio of the average rupture velocity to the subevent rupture velocity. The shear waves radiated by eight moderate (3.6≤ML≤4.9) multiply recorded aftershocks of the 1975 Oroville, California, earthquake are analyzed using measurements of the corner frequency, the characteristic frequency, the pulse duration, and the signal duration. Comparison of these four different estimates of source size indicates that the corner frequency and the characteristic frequency are more sensitive to the dominant subevent size than to the overall event size. The difference between the estimates of source size determined from the signal duration measurements and the spectral measurements increases as the rupture complexity increases.