We explore the nature of systematic errors that can arise in measurement of black hole masses from single-epoch spectra of active galactic nuclei (AGNs) by utilizing the many epochs available for NGC 5548 and PG1229+204 from reverberation mapping databases. In particular, we examine systematics due to AGN variability, contamination due to constant spectral components (i.e., narrow lines and host galaxy flux), data quality (i.e., signal-to-noise ratio, S/N), and blending of spectral features by comparing the precision and accuracy of single-epoch mass measurements to those of recent reverberation mapping studies. We calculate masses by characterizing the broad Hbeta emission line by both the full width at half maximum and the line dispersion and demonstrate the importance of removing narrow emission-line components and host starlight. We find that the reliability of line width measurements rapidly decreases for S/N lower than ~10 to 20 (per pixel) and that fitting the line profiles instead of direct measurement of the data does not mitigate this problem but can, in fact, introduce systematic errors. We also conclude that a full spectral decomposition to deblend the AGN and galaxy spectral features is unnecessary except to judge the contribution of the host galaxy to the luminosity and to deblend any emission lines that may inhibit accurate line width measurements. Finally, we present an error budget which summarizes the minimum observable uncertainties as well as the amount of additional scatter and/or systematic offset that can be expected from the individual sources of error investigated. In particular, we find that the minimum observable uncertainty in single-epoch mass estimates due to variability is ~ 20 per pixel) spectra.