There isn't a simple/short answer to this question. You can find information on different peak shapes in various text books (e.g. the Billinge/Dinnebier text or the text on Rietveld Refinement by me, Robert and Andreas).
The appropriate choice of peak shape will depend very much on the information you're trying to extract from your fitting. In some cases it might be critical to describe the peak shape "perfectly" (e.g. if you're trying to extract information on microstructure or faulting). In other cases it might not matter if the peak shape isn't perfect. For example, if you're doing structural work and you have the same misfit on all the reflections then it probably won't affect your extracted coordinates. Improving the peak shape model might lower your Rwp, but won't necessarily give you better information. If you're just trying to extract accurate cell parameters and the experimental peaks are symmetric you'll probably get the same values with more or less any symmetric peak shape model. However if the experimental peaks are influenced by factors that lead to asymmetry and peak shifts (axial divergence, transparancy, etc) then you would need to describe that appropriately.
Hope that's some help.