Essentially, the goal of FMEA is to identify each possible failure mode and the possible consequences of that failure. For each failure mode identified, a Risk Priority Number is assigned. An RPN is a way of ranking the most critical failure modes to address.
S = Severity (1 = no effect, 10 = worst possible case with no warning)
L = Likelihood (1 = not going to happen, 10 = almost certain)
D = Detectability (1 = easily detectable, 10 = no detection)
The scales for these quantities are largely trivial, as they are designed to be relative within a particular design. As long as they are consistent throughout the FMEA, the highest RPNs will correspond to what should be the highest priority.
Once RPNs have been determined, it gives the designer an obvious priority list for where to look at improving reliability in the circuit. Efforts should be made to reduce high RPNs, and lower RPNs can be ignored (until the higher RPNs have been addressed anyway...)
A more detailed description of FMEA, as well as a sample worksheet and suggested scales for S, L, and D is provided in Torsten Lehmann's lecture notes. These can be found at http://subjects.ee.unsw.edu.au.