A new criterion for the "useful" resolution of electron micrographs of macromolecular particles is introduced. This criterion is based on estimation of the spatial frequency limit beyond which the spectral signal-to-noise ratio (SSNR) falls below an acceptable baseline. Applicable to both periodic and aperiodic specimens, this approach is particularly apposite for sets of correlation-averaged images. It represents a straightforward and intuitively appealing generalization of the traditional method of estimating the resolution of crystalline specimens from the spectral ranges of periodic reflections in their diffraction patterns. This method allows one to assess how closely the resolution of an averaged image based on N individual images approaches the ultimate resolution obtainable from an indefinitely large number of statistically equivalent images. Inter-relationships between the SSNR and two other measures of resolution, the differential phase residual and the Fourier ring correlation coefficient, are discussed, and their properties compared.