In the field of electronic or computerised control apparatuses, known methods of compensating for disturbance signals rely on less than ideal analytical or empirical models. There is provided a controller (200) or method of controlling which observes and learns the correlation between various measured signals and automatically learns how to control the apparatus. It has a primary input (x.sub.R) for the measured state of the apparatus, and signal processing means (604, 606) responsive to the input signal for generating a control signal (V) to maintain the apparatus in a desired state, the controller (200) having further input(s) (X.sub.1, X.sub.n) for additional measurements of the apparatus or its environment, the signal processing means using a weighted average (606) of results of different fixed impulse responses (611 615) to each input to modify the control signal, and means (630, 800, 802) for conditioning the response automatically in response to temporal cross-correlation observed between the measurement signal(s) and the control signal. The controller can be pre-taught with an appropriate response, or it can learn it during initial operation.