The ensemble methods show the ability to obtain the geometry of natural fractures in high-permeability reservoirs. However, hydraulic fractures in unconventional reservoirs are different and the ability of these methods to invert production data to determine hydraulic fracture geometry is unknown. In this work, Embedded Discrete Fractures method is selected as the simulation method. The EnKF and EnRML methods are chosen as the history matching methods. Several simulation cases are used to test these algorithms’ performance in both conventional and unconventional reservoirs. Also, a novel gradient history matching process has been proposed to solve part of the problems. By comparing history matching results of different cases, Ensemble methods’ matching quality is very sensitive with prior information. Production of wells in low-permeability reservoirs is not sensitive with the unconnected fractures, which will lead to multiple minima in the objective function. It also causes big difficulties in the estimation of the gradient for ensemble methods. Thus, it is even more important to add prior information properly in the unconventional reservoirs’ history matching, which can narrow the problem dramatically. The potential and limitation of using a gradient method to solve this kind of problem are also shown in this paper.