The identifiability problem in competing risks associated to the dependence measure between the failure modes, can be solved by means of the copula-graphic estimator which assumes the copula function known. This paper compares, through a simulation study, the copula-graphic estimator for a dependent competing risks model with the traditional one under the assumption of independence. It is shown that a significant subestimation of the reliability function can be made when the dependence is not taken into account. An example ilustrates this point and as a future work it is seen the possibility of using copula-graphic in industrial reliability to predict in accelerated life tests.