I use Poisson sigma-reject on small stacks (4 to 8 subs). Its very reliable at detecting outliers after good normalisation (selecting both background and highlight manually).
For large stacks (8+ subs), STD sigma-reject is preferred. Poisson can still be used however but I've found that you don't need a strong standard deviation value.
If the data is of good quality a deviation value of 2.2 works well. I'll drop down to 2 or 1.8 for mediocre. For a small stack of RGB, at times I'll drop down to 1.5 deviations. I rarely go lower than this.
You may be in for a surprise. With poor data normalisation, Fred V may not be computed as an outlier thus not rejected. As a result Fred would be embedded to increase 'signal over noise'. Put another way, he may be full of it, but his parted wisdom is 'gold over crap'.
Thanks everyone, general consensus seems to be heading the same way - I have been using roughly the same percent rejection however always used poisson reject for all my stack rejection.
