This paper presents a review of the g-h distribution in operational risk and proposes a modification to the method developed by Hoaglin (1985) to estimating the parameters. The modification consists in the estimation of the parameter h using a robust regression. We estimate OpVaR by g-h and POT methods in two applications. The results show that the g-h method is useful in operational risk, but care must be taken when the distribution of losses presents extremely heavy tails.