[quote author="Brian Roth"]OK....
[snip]
In my experience, and as a rule of thumb, the tranny's VAC rating needs to equal the desired VDC output (after regulator) to accomodate for diode voltage drops and ripple on the filter cap.
Bri[/quote]
I agree that's a pretty good rule-of-thumb for voltages in this range. At higher voltages it can be relaxed a bit, and at lower ones it cuts it a little too close (for example a ~6V filament supply), since the diode drops and regulator differential take out a bigger fraction of the voltage.
For Fairchild's 7818 they rate the output under load to 1A with a 3V drop (21V in), although the Dropout Voltage as an input-output differential is given as 2.0V typical. For the first number with 21V in, the output is allowed to sag to as little as 17.1V, and can be as high as 18.9V.
[quote author="buttachunk"]Sorry I was unclear,, the traffos were rated 12VAC, dual secondary,,, they ended up around 16-0-16 with no load... I thought that there was a margin of error of like 3-4 volts under with regs...
Ended up designing and building a little charge-pump, which boosted up to +-20.5-ish volts (per rail). I could have fed the regs with this, but ended up bypassing them and went through 10r resistors instead. The rails are now at +18.3VDC and -18.4VDC under load, measured at the opamp pins,, running far quieter than the reg'd supplies....[/quote]
Linear regs can only reduce their input voltages in order to give one a regulated output voltage---all they are doing, after all, is putting the equivalent of a variable resistance in series with the raw supply. There are some varieties of switching regulators that can both boost and cut as required, but typically even for them it's just one or the other.
Most op amps are fairly tolerant of power supply fluctuations at least at low frequencies, although they can fall apart badly at higher. The reason to use regulators is often more one of providing very low ripple without a lot of passive filtering, rather than tight control of the d.c. output per se.
But in systems where d.c. levels, and not just their ratios, are important, regulators are essential. Many times better system design can make the power supply sensitivity a lot less, but decent global regulation may still be the sensible way to go.