In nature, most of the surfaces are rough and waveguide flanges are not exception. And these surfaces topography are of high importance in the response of the waveguide flanges connection system. These surfaces topography are usually described by statistical parameters. And statistical parameters depend strongly on the resolution of the roughness-measuring instrument, and hence the values of statistical parameters are not unique for a surface. As a result, the predictions of passive intermodulation (PIM) level of microwave devices based on these statistical parameters also may not be unique to a pair of contacting surfaces. Fortunately, fractal method is independent on the resolution of the roughness-measuring instrument. Therefore, the randomness of the rough surface is not affected by the differences in the model. Consequently, if PIM level of rectangular waveguide is predicted with fractal parameters, the predicted value will be unique. Based on this motivation, a new model is developed. In this model, the fractal method is used to characterize the surface topography of waveguide flanges. Then, the third order PIM level of a rectangular waveguide flange is predicted. The analytic results are verified by PIM test set-up. This work is help to understand PIM phenomenon and design the low-PIM waveguide.