I consider copula-based double-hurdle models with flexible marginal distributions. While the copula must be specified in advance, the marginal distributions are assumed to belong to a quite general class of distributions and need not be specified by the researcher in advance. A simulation study indicates that the copula-based double-hurdle models with flexible margins perform well even when the selected copula has been misspecified. An empirical application shows that copula-based double-hurdle models with flexible margins may outperform the conventionally used classical double-hurdle model based on a bivariate normality assumption.