Andreas Your questions sounds like it can better be answered by someone who's good with statistics, that's not me. One minute averages should work OK for the most part. At that update rate, it is harder to tell the difference between LF white noise and Bad step noise. But if you can not see it, then it is not likely a limiting factor. Can use digital or analog filtering, anything that lets you see changes of differences in the sub PPM. The bad type noise, called popcorn and others, is not all that random, It looks more like small steps. What I use for grading is the size of the 'typical' low freq steps and how often they happen. In a good references the step sizes will be under 0.1 PPM with a few per hr, poor ones can be 1 or 2 PPM, and as often as a few times per minute. example: Attached is a 16 hr strip chart of my zener plug. It shows about 0.1 PP noise and about 3 steps per hr. The 845 was set to 3uv FS on the noisy 5 hr center plot, and at 10uv FS on the rest. Vertical is 1 hr per division, updated at 1 per second thru a 15 second filter. Near the bottom of my 'going to get around to it list' is to plot the ADEV of different voltage references for a few days using a tau zero of 1 sec. I think that would be a good way to show voltage reference performance nicely out to hrs, and not be so subjective. ws **************** Hello Warren, thanks for your response. I had a look on the manual of the Fluke 845: nice part. But: For the first evaluations of low frequence noise I will have to use my 24Bit ADCs. So without larger changes on my software I will have to use the 1 minute averaged values. What would be better for comparison of the low frequency noise: Summing up the absolute values of the differences between the 1 minute values and calculating some average change per minute on the values. Or: calculating some "effective noise value" on the AC-part of the reference voltage. With best regards Andreas