I suggets the following:

(1) Use your code to caculate a reduced case when there is a known exact solution and compare with it. For example, in structural analysis, there are exact solutions for deflection of some bams, plates and shells.

(2) Use your code to caculate some complicated demo problems available in some commercial software packages and compare with your code.

(3) Use engineering judgment or secientific understandng to assess your results.

These should help you gain confidence in your code but in theory there is no guarantee a complciated code would be completely error free.

Thank you Ouyang.

But you have mentioned as code many times what does it mean actually ?

Muhammad your saying that it largely depends on failure. But i want to know how will we judge that failure as occurred in any analysis software from the values (max & min) we got if we are not having any related experimental data.

Thank you Muhammad.

Failure in FEA need to be properly defined. I am assuming you are using a commercial FEA package such as ANSYS. Normal prerequisite should be that you should have some kind of an idea about the first order results to expect from the analysis. When you get the results, compare the results with the expectations, if they are close enough, then you are fine. If the results are way far removed from expectations by some order of magnitude such as 10^4 etc., then you know something wrong happened during the meshing or analysis stage.