If you mean, of an arbitrary, truly general circuit, there need not be any single figure, nor does it need be computable at all!
Even for simple circuits, it can vary with voltage, time, internal configuration or state, etc. And given those considerations, there are many (also non-unique) methods of averaging it, to get an equivalent value.
A classic example being an RF amplifier: the output capacitance of a transistor varies with voltage (both the supply or bias voltage, and the signal voltage), so that the matching impedance varies with voltage as well. The best you can do is match the average (given a suitable averaging method), but even that is not very helpful, because it is only valid for one particular level.
If you have a fairly well-behaved, but otherwise unknown, circuit in a box (a "black box" experiment), and wish to measure its characteristics, then you can use a number of theorems to find this: you can use different load resistors or impedances to calculate the Thevenin equivalent, or the Norton equivalent, or a number of matrix-oriented methods that arise from the study of simple, abstract amplifier circuits or RF systems (e.g., the h, z and s parameters).
Tim