I have done some FMEAs.  I hear I do them backwards.  I have discovered several interesting problems with systems and circuits with my method though.  Some of my system engineering friends say what I do is closer to a Fault Tree Analysis.  That may be. 

What I do is to assume a failure mode for each component in a circuit.  I then look at that fault and see what the circuit does (given that fault).  I then do the standard FMEA thing to the resulting list.  That is assign a probability number to the fault , assign a detection number (will the user understand that the output is a result of a fault condition) and a severity for the resulting system problem.

The probability number represents the chance for something actually happening the way the failure mode I am looking at describes.  For instance, will a NPN transistor really fail collector emitter short and base emitter open.  Not often.

The detection number represents the chance that a the user of a device will recognize the devices response is the result of a failure of some sort.  For instance the output of a thermometer might give you a anwer of 36 degrees when the temperature being measured is 38 and one segment of the LCD is out.  That is hard to detect.  When the anser is 3d in the same circumstance, it is easier to detect.

After the severity for the system responce to the failure is assigned a number, the three numbers are multiplied.  These numbers represent a prioritized grouping of the component level faults.  In many organizations there are procedures defining the level of concern that each of these numbers represents.

If this is an exercise that you do, let me look at your circuit and I will help you with the failure mode analysis.