Comparison with experimental data or theory is definately the way to go in validating any models. However, such data is not always available. There are a number of basic checks that you can do:

1) displaced shape (if ther is one) is it sensible or what you might expect?
2) do the reaction forces and inertia forces equal the applied loads?
3) look at the junction between elements and see if there is a significant discontinuity in temperatuers, forces or stresses etc.
4) Do some basic calculations to get upper or lower limit bounding values at suitable locations - compare to the FEA results.

Often it is sensible to start with a simplified structure or a benchmark case to demonstrate that the methods/options used are suitable.
Finally it is best to get someone else to review your approach, model and results as it is only to easy to make mistakes.

TERRY

Terry,

thanks for the input, it's quite helpful. I think of some other basic principles, like energy conservation, hourglass, etc. And if possible to refine the mesh, check if the max. stress/strain converge or not.

Thanks again,

Lake

Terry's recommendations are good ones.  This is a hard question to answer generally, especially for explicit solvers. I especially like Terry's recommendation #3.
Brad