The noise, due to statistical fluctuation of signal electrons, determines the ultimate resolution limit of imaging and analysis. The signal electrons (N) for each pixel are formed through two stages of statistical processes; Poisson distribution for the number of incident electrons (No) and polynomial distribution for forming signal electrons. The signal electrons therefore obey the Poisson distribution of average nOp (no:average of incident electrons, p:probability of forming signal electrons) and of standard deviation . In order to improve the resolution limit set by the statistical noise of signal electrons, a new method of normalization is proposed for STEM.In STEM imaging of thin specimens, the total transmitted electrons correspond to the incident electrons. It is therefore possible to know the incident electrons by summing up all the transmitted electrons and to normalize the signal electrons by incident electrons, for eliminating the influence of fluctuations of incident electrons. The average E and the standard deviation σ of normalized signal electrons are expressed by p and , respectively, for no> > 1 , indicating that E/σ value is improved by a factor of .