A characteristic energy of individual gamma-ray burst (GRB) spectra can in most cases be determined from the peak energy of the energy density spectra ($\nu{\cal F}_{\nu}$), called '$E_{\rm peak}$'. Distributions of $E_{\rm peak}$ have been compiled for time-resolved spectra from bright GRBs, and also time-averaged spectra and peak flux spectra for nearly every burst observed by CGRO-BATSE and Fermi-GBM. Even when determined by an instrument with a broad energy band, such as GBM (8 keV to 40 MeV), the distributions themselves peak at around 240 keV in the observer's frame, with a spread of roughly a decade in energy. $E_{\rm peak}$ can have considerable evolution (sometimes greater than one decade) within any given burst, as amply demonstrated by single pulses in GRB110721A and GRB130427A. Meanwhile, several luminosity or energy relations have been proposed to correlate with either the time-integrated or peak flux $E_{\rm peak}$. Thus, when discussing correlations with $E_{\rm peak}$, the question arises, "Which $E_{\rm peak}$?". A single burst may be characterized by any one of a number of values for $E_{\rm peak}$ that are associated with it. Using a single pulse simulation model with spectral evolution as a proxy for the type of spectral evolution observed in many bursts, we investigate how the time-averaged $E_{\rm peak}$ emerges from the spectral evolution within a single pulse, how this average naturally correlates with the peak flux derived $E_{\rm peak}$ in a burst and how the distribution in $E_{\rm peak}$ values from many bursts derives its surprisingly narrow width.