Any FEA is performed based on assumptions. The more assumptions, the errors possible. The "trick" is to understand the assumptions being made and their possible impact on the resulting analysis. This is true of any FEA program.
As for CosmosWorks specifically...it does a reasonable job and is as accurate as the underlying geometry and assumptions. I as suspicious of the default mesh sizes and like to vary the mesh size to see convergance, and to identify areas that really need custom meshing. So, while C/Works is easy to use, don't sign-off on the first set of numbers it spits out. Verify, verify, and verify again.
.....Peter.....