Updating a non-linear discriminant function estimated from Gompertz populations is investigated. The updating procedure is considered when the additional observations are mixed or classified. Using simulation experiments the performance of the updating procedures is evaluated via relative efficiencies. On the other hand, the asymptotic expectations of the total probabilities of misclassification for mixture and classified discrimination procedures are evaluated. Then the asymptotic efficiency of the mixture discrimination procedures relative to the completely classified are obtained and discussed for some combinations of the parameters.