Let M (x) be a regression function which has a maximum at the unknown point θ. M (x) is itself unknown to the statistician who, however, can take observations at any level x. This paper gives a scheme ...