The accuracy of progeny testing for a continuous trait when the auxiliary information was based on ‘all-or-none’ type of traits with a multi-factorial threshold model was examined. The accuracy of the progeny test was found always to increase with the inclusion of the auxiliary trait. It followed the same pattern as in the case of continuous auxiliary trait. However, there was an additional feature in that the accuracy was also dependent on the probability of incidence of the discrete trait unless either the progeny group size was very large, and/or there was no genetic correlation or else the probability of incidence was itself extremely low. Compared with the case when the auxiliary trait was continuous, there was a loss in accuracy which can be looked upon in two ways. First, the continuous auxiliary trait is itself made ‘all-or-none’ type by a threshold and the accuracies compared in the two cases. The loss increases symmetrically in either direction with increase in positive or negative directions of the difference in genotypic and phenotypic correlations. Secondly, a comparison can be made of the accuracy when the auxiliary trait is normally distributed with that when it is binomially distributed but with the same amount of genetic information on the observed scale as given by the heritabilities and genetic correlations. No loss now occurs if the two classes of the binomial trait are equally likely and the loss in accuracy increases with the decrease in the standard deviation of the binomial distribution for given values of other parameters. It was found also that the use of binomial auxiliary trait reduced the number of progeny required to attain a pre-assigned level of accuracy, resulting in a decreased cost of the programme compared with that when no binomial auxiliary trait was used.