We discuss the advantages and disadvantages of estimating the expected shortfall of financial asset returns using seminonparametric distributions instead of more traditional parametric ones. More specifically, we focus on two seminonparametric distributions that are based on expansions of the normal distribution and have recently received attention in the financial econometrics literature: the fourth-order Cornish-Fisher and Gram-Charlier distributions. The performance of these distributions is compared to that of other popular parametric distributions in this context: normal, Hansen’s skewed t and Johnson distributions. We estimate the expected shortfall of a wide variety of financial series (including several market indexes, euro to US dollar exchange rate, bitcoin and various stocks) under these distributions, specifying a non-linear Garch model for the conditional variance. The performance of the estimates is compared by means of backfitting tests and some other relatively standard selection criteria. In terms of backfitting, all models except the normal perform adequately in most cases. However, the other selection criteria that we use suggest that the seminonparametric distributions, especially the Cornish-Fisher one, outperform the traditional parametric distributions.