A technique based on analysis of the cross spectra of single-use data from pulsars is described which can be used to measure the coherence properties of, and search for coherent periodicities in, pulsar signals. Several simulated pulsar models are constructed and analyzed to test the technique. From an analysis of 2000 consecutive pulses from PSR 2016 + 28 it is concluded that the micropulse coherence time of 430 MHz is less than the spin period for this object. This could could be consistent with a coherent model if the random phase jitter in such a model is 35 percent or more of the micropulse quasi-periodicity. The subpulse separation of roughly 10 ms is coherent across 4-5 pulse periods and is reasonably consistent with a model of subpulse drift that allows for variation about some mean drift rate. A two-component shot-noise model can reproduce much of the structure seen in the individual pulses as well as the spectra of PSR 2016 + 28.