In this study, the newly proposed numerical models have been verified and used to characterize the fracture distributions in a naturally fractured reservoir conditioned to tracer transport behaviour. The stochastic fracture modeling approach is implemented to generate the randomly-distributed natural fractures which are dealt with the embedded discrete fracture model (EDFM) while ensuring its sufficient accuracy. To be specific, the matrix domain is discretized using the structured grids, within which each embedded fracture is divided into a series of segments. Subsequently, non-neighbouring connections allow us to couple the flow of fluid and tracer between the non-neighbouring grid cells, while the historical tracer profiles are matched to delineate the geometry and properties of the fractures by taking multiple tracer transport mechanisms into account. Furthermore, the influences of fracture number, fracture length, fracture orientation, and tracer dispersion on the tracer production concentration have been investigated through sensitivity analysis. The response of an interwell tracer model is sensitive to the fracture parameters rather than tracer properties. A fracture network with its orientation parallel to the mainstream direction will cause the earliest tracer breakthrough. The tracer breakthrough time with an average fracture length equal to 40 m is 110 days earlier than that with a mean fracture length value of 10 m, while the tracer production peak concentration for the former is nearly two times higher than for the latter. A larger fracture number results in an earlier tracer breakthrough and an intermediate fracture number will lead to the highest tracer production concentration. Additionally, the newly developed model has been validated through its comparison with the commercial ECLIPSE simulator and then extended to field applications to identify the possible fracture distributions by simulating multi-well tracer tests in the Midale field. The flexible and pragmatic EDFM-based method developed in this study can model the interwell tracer flow behaviour as well as characterize the properties and geometries of the natural fractures with better performance on accuracy and calculation efficiency in comparison with other fracture simulation methods (e.g., local grid refinement (LGR) method).