Estimation of small signals from counting experiments with backgrounds larg
er than signals is solved using maximum likelihood estimation for situation
s in which both signal and background statistics are Poissonian. Confidence
levels are discussed, and Poisson, Gauss and least-squares fitting methods
are compared. Efficient algorithms that estimate signal strengths and cofi
dence levels are devised for computer implementation. Examples from simulat
ed data and a low count rate experiment in nuclear physics are given. (C) 1
999 Elsevier Science B.V. All rights reserved.