In real problems, it is usual to have the available data presented as intervals. Therefore, different approaches have been proposed to obtain a regression model for this new type of data. In this paper, we represent the interval-valued response variable Z = [Y L , Y U ] as a bivariate random vector and we consider the copula's theory to propose a general bivariate distribution for Z, creating a more flexible random component to the model. Inference techniques and a residual definition based on deviance are considered, as well as applications to synthetic and real data sets that demonstrate the usefulness of the proposed approach. The new method is also compared with other methods reported in the literature.