Unfortunately jobs seem to execute differently in some instances - bit like triggers.

Saying that though most times an the execution will continue after an error - even if the job does end up as failed.
What is the error that you are trying to trap?