Naturally fractured reservoirs (NFRs) gain much attention worldwide because they are often encountered in aquifer remediation, CO2 sequestration, and hydrocarbon extraction. In hydrocarbon extraction, however, oil recovery by gas injection in NFRs is usually low, because of poor sweep efficiency. During gas injection, the displacement front is unstable. Conformance problems, such as gravity override, viscous fingering, and channeling, take place because the gas has a lighter density and lower viscosity compared to reservoir fluids, and tends to flow preferably through high-permeability zones in heterogeneous reservoirs. In addition, open fractures can have much greater conductivity than the matrix. As a result, gas flows through fractures, leaving much of the matrix unswept. Foam, by adding surfactant solution to gas injection, can effectively mitigate conformance problems by greatly reducing the mobility of gas. During foam flooding in porous media, the displacement front is more stable, and more gas is diverted to unswept zones, hence improving the sweep and increasing oil recovery. Foam can also be created in fractures, where it builds up a viscous pressure gradient and thus diverts the flow of gas into the matrix. As a result, the sweep is improved. In the field, foam pilots have achieved an increase in oil production rate and a reduction in gas/oil ratio. Despite this success, foam application in NFRs is still much less understood than in unfractured porous media. In this dissertation, we aim to expand our understanding of foam in fractures through an experimental approach. To this end, we create four 1-m-long, 15-cm-wide glass model fractures (Models A, B, C and D) with different roughness and hydraulic apertures. Each model consists of two 2-cm-thick glass plates. The top plate is smooth and the bottom plate is roughened on the side facing the top plate. Between the two plates is a slit-like channel representing a single geological fracture. Model A has a roughened plate with a regular roughness. Models B, C and D, with increasing hydraulic apertures, use the same roughened plate with an irregular roughness. We profile the roughness of the roughened plates and study the aperture distribution of the model fractures to characterize the geometry of the model fractures. With local hills (maxima of height) and valleys (minima of height) on the roughened plates, the distribution of aperture of model fractures can be represented as a 2D network of pore bodies and pore throats. In the experiments, we inject pre-generated foam into the model fractures. We study foam behavior after foam flow reaches steady-state. As our models are transparent, we use a high-speed camera to directly visualize and record images of foam in the model fractures. Using ImageJ software, we analyze foam images to quantify the properties of the foam.