Sigma combines CCD FITS images in a way that reduces noise in your combined CCD image compared to using a simple average/mean combine. For instance I'm sure you've seen cosmic ray hits and dark/light spots from hot pixels on some of your images. If you sum or average 5 images with these types of noise the noise follows into the resulting image. However, if you did a median combine much of the noise is gone BUT the overall signal/noise of the image is reduced. That is the median combined image looks grainier than the sum/averaged combined image.
Sigma has two algorithms to reduce noise when combining images. The classical sigma rejection algorithm and my own Standard Deviation Masking algorithm. The results are quite good with either algorithm and you can compare them to the mean/median with a button click.
Some caveats though... if you are combining light images they must be already registered (aligned) with respect to one another. Sigma cannot register images at this time however programs like MIRA, MaximDL, Registar, and CCDSoft can register light images. Darks and Flats do not need to be registered nor should they. Also, the more images you combine the better Sigma will work. To truly see the best results you should combine at least 10 darks, flats or light images.