Recently, the usage of optical fiber coils has increased significantly, especially in the design of physic and chemical sensors. Therefore, it is important to test the theoretical current models developed to predict the power loss throughout optical fiber. In this paper a pioneer and popular model, the Marcuse model of power loss, was studied and evaluated for optical fiber coils of small radii. Power attenuation in a bent fiber data was collected using an Optical Time Domain Reflectometer (OTDR), and it was compared to the theoretical predictions of the Marcuse model. It was observed that the model predicts correctly the attenuation behavior for usual curvature radii, however, it fails to predict accurately the attenuation behavior for small curvature radii, underestimating considerably the actual power loss. Also, it has been observed that at small radii the power loss parameter 2α and the mode propagation constant of the wave guide β stop being constants and become functions of the optical path, particularly of the number of loops in the coil. It is possible that new mechanisms of light leaking are present, due to the extreme distortion of the modes configuration into the fiber at small radii. Those mechanisms cannot be described by a model that considers a power loss parameter 2α, and more specifically the mode propagation constant of the wave guide (β) as constants. Then it is important to develop other models where the previous parameters can be considered as functions of the optical path.