Abstract Methodologies and numerical tools are available (1) to construct geologically realistic models of fracture networks and (2) to turn these models into simplified conceptual models usable for fieldscale simulations of multiphase production methods. A critical step remains however, that of characterizing the flow properties of the geological fracture network. The multiscale nature of fracture networks and the associated modeling cost impose a scale-dependent characterization: (1) multiscale fractures that may be characterized in local dynamic test areas, e.g., drainage areas involved in well tests, through the calibration of geologically realistic fracture models; and (2) large-scale faults that are characterized through reservoir-scale production history simulations that involve upscaled flow models with an explicit fault representation. However, field data are commonly insufficient to fully characterize the multiscale fracture properties. Therefore, efficient inversion methodologies are necessary to sample wide ranges of property values and to characterize a variety of solutions, i.e., fracture models that are consistent with dynamic data. This article presents an inversion methodology to facilitate the characterization of fracture properties from well-test data. A genetic optimization algorithm has been developed and coupled with a three-dimensional fracture model upscaling simulator to perform the simultaneous calibration of well-test data, i.e. equivalent transmissivities K· h, with K the equivalent permeability that takes into account fracture flow properties, and h the reservoir thickness over which the well test has been interpreted. Several genetic crossover and mutation strategies were studied and tested on three geologically realistic fractured reservoir models, involving both small-scale diffuse fractures and large-scale sub-seismic faults. The characterized diffuse fracture properties are mean length, mean conductivity, orientation dispersion factors, and facies-dependent properties such as fracture density. The fault network conductivity is also characterized. The effectiveness of this inversion methodology to characterize physically meaningful and data-consistent fracture properties is discussed.