In EMAN 1.8 I run a refinement, and when I look at the number of particles in my class averages I get really weird values that obviously aren't correct.
It appears that there is a bug with the "median" option in EMAN 1.8. It is fixed in current snapshots. If you are using the 1.8 release without CTF correction, simply omit the 'median' option. This will perform an average rather than a median, which is fine. The only thing to be aware of with this, is if you are using CCD image data, you need to make sure you have removed x-ray pixels before doing your reconstruction.