In this communication we present an empirical approach to the task of
estimating the reproducible limit of resolution that can be achieved w
hen a set of images is averaged. The proposed solution is based on a r
andom subsampling method. We derive confidence limits for two of the m
ost widely used measures of reproducibility in this field, the differe
ntial phase residual (DPR) and the Fourier ring correlation (FRC). Res
ults both for negatively stained specimens and for those embedded in i
ce are presented. Our results provide a common framework that allows a
number of previous observations and suggestions regarding these measu
res to be further analyzed.