I don't understand why the measured level will vary with the total
input - surely provided the A/D converter is not overloaded, the
signal component being displayed should only be proportional to
the amplitude of that component of the input signal?

Thanks to all who answered so far (I am still open to suggestions !).
Jim, yes, I used the wrong phrasing to express what I had in mind.
What I meant is that I cannot use the instantaneous level in the time
domain as a parameter of the equation that computes the dB value
of the bin, as the same input level can be obtained by summing
different frequency components.

Would the following method be acceptable :
I compute (using the Parseval equation) what the single bin magnitude
would be if I feed to the ADC only a pure sinusoid such that the ADC is
just at the saturation point (i.e. a level of 32767 for a 16-bit card).
Then I use the computed magnitude as a reference value to display
the dB values of the various bins, which would of course be all
negative with this method.
What do you all think of this ?  Thanks for feedbacks.

73  Alberto  I2PHD