Using a failure criterion such as the shear or tensile failure keywords should be considered carefully. There are many poor uses of this feature... I would suggest reading the keyword manual in abaqus to understand if you want element deletion on or off. Also look in literature for how other researchers have gone about modeling your type of simulation. For instance sometimes the parameter is tuned to deal with mesh distortion issues.
I would also caution that failure criterions are not very robust. For instance failure does depend on temperature, strain-rate, loading history, microstructure, etc... So if you have a ballistics simulation for instance, you would most likely need experiments done at these extremes to get the appropriate model parameters.
If the answer is yes, element deletion is needed, then check out this primer which schematically shows the identification of the failure parameter (UTS value as the onset of damage):
http://imechanica.org/files/l9-damage-failure.pdf
Find a good dataset in literature for your particular material and identify the parameter. If you have tensile curves and you might assume an isotropic material and use the von Mises relation between stress and shear (shear=1/sqrt(3)*tensile_strength. Whether this is a good assumption depends on the forming and thermal history of the material you are simulating.