My preference for this is to use Standard Deviation. At +/- 2 standard
deviations you will be excluding the 5% of the data that most significantly
varies from the mean. That removes the big anomolies but leaves the majority
of the data in pool.
--
HTH...

Jim Thomlinson


"Muk" wrote:

Hi,

I am dealing with sets of real data and as would be expected there are lots
of 'noise'. I would like to set a baseline so that calculations are
performed only on the data above the baseline. I know that the kurtosis
function shows how "peaked" the data is but I am looking for a data sieve
that ignores the random noise and accepts the real data (say anything five
times higher than random noise).

Thanks