Based on a random sample of individuals, new conventional formulae for estimating gene frequency in the ABO and the ABO-like system are presented. Applying the formulae on a large number of samples, it is found that the new method is simple and yields estimates as accurate in at least two decimal places as the maximum likelihood solutions. The method however might be fraught with bias when the O phenotype frequency was low or even absent. Degrees of this bias have been discussed in terms of the number of iterative processes by a method of gene counting and of a goodness of fit to the Hardy-Weinberg proportions.