The character of detector noise is explored in photometrically corrected images from the short-wavelength and long-wavelength prime intensified vidicon cameras of the International Ultraviolet Explorer A protocol is proposed for deriving realistic "noise models"-crucial to the application of Optimal extraction algorithms like that of Kinney, Bohlin, and Neill (1991, PASP, 103, 694)-from the available collections of UV-Flood calibration images. The protocol includes evaluation of the "noise-filtering" properties of the SWP and LWP cameras through 2-D spatial power spectrum analysis. The two vidicon cameras behave nearly identically. For both, the incomplete removal of the pixel-to-pixel sensitivity pattern can lead to a factor of up to two enhancement in the apparent noise, depending on position in the image. Even with good suppression of the pixel granularity, however, the remaining random noise can exhibit saturation behavior that causes the S/N to cease improving with increasing exposure. The random noise itself exhibits a two-component character: a normal white-noise field superimposed on a filtered (Gaussian-like smoothing) background. The influence of the smooth component varies strongly with position. Nevertheless, when all of the relevant effects are considered, the underlying "pristine" noise models show essentially no dependence on spatial position, except for an unusually noisy patch on the LWP camera. Two additional sources of noise, beyond the largely photometric contributions documented above, are microphonics and cosmic particle radiation. Microphonics are important in only a few exceptional circumstances, but cosmic ray "bright spots" set an effective limit of ~ 4 hr on useful SWP-LO exposures of (unresolved) emission-line objects, even those conducted during low-radiation time.