The results of a simulation study comparing the method of maximum likelihood for binary regression with unreplicated data and two approximate methods are presented and discussed. The two approximate methods are that of Cox [1966] and unweighted least squares with the 0's replaced by -3 and the l's by +3. The results indicated that with 20 or more data values the maximum likelihood methods worked quite well but that with fewer than 20 the approximate methods are to be preferred.