Two-phase (air-water) flow experiments were conducted in artificial horizontal fractures (narrow channels). Two experimental set-ups were utilized. One set of experiments was performed by using two glass plates (1 x 0.5 m) with a gap width of 1 mm. The second set of experiments was performed using two bricks made of baked clay (28 x 14 cm) for which three gap widths of h(1) = 0.54 mm, h(2) = 0.40 mm and h(3) = 0.18 mm have been tested. Air and water were injected separately, through alternating capillary tubes for the first set-up and through a porous medium for the second. For each experiment, the fracture was initially saturated at constant water flow-rate, and air injection was then started. When steady state was reached, pressure drop and liquid volume fraction were measured. Then, air injection was increased stepwise and the experiment was repeated at different liquid flow rates. By varying the flow rates of each fluid phase, different flow structures were observed for the glass channel experiment : bubbles, fingering bubbles, complex, annular and droplet flow. These flow structures show more similarity to those observed in pipes than to those expected in porous media. Using the formalism developed for two-phase how in pipes, and by taking experimental observations into account, a theoretical relationship for the two-phase pressure gradient is proposed. This relationship is evaluated with experimental data. Then, the results are analyzed with three models. First, the Lockhart & Matinelli model gives a good fit for both pressure drop and liquid volume fraction against the Martinelli parameter. Second, by considering the two phases flowing in the fracture as a single phase with averaged properties, the appropriate friction factor and Reynolds number of the mixture are defined. This model, which is similar to the homogeneous model, permits the selection of the experimental data corresponding to laminar flow. Finally, by using the generalized Darcy model, it was found that for laminar flow, the liquid-phase relative permeability is equal to the liquid volume fraction, while the gas-phase relative permeability is not a linear function of the liquid volume fraction.