The transition features on the variance fractal dimension trajectory are investigated using a noise separation technique for distinguishing various seismic refraction signals from noise. The seismic refraction signals are characterized by short-period vibration waves and include primary body (P) waves, both refracted and reflected from subsurface interfaces, as well as random noise and coherent noise such as diffracted and scattered waves. A window size of forty-eight samples and window spacing of eight-sampling intervals were chosen and tested to calculate the variance fractal dimensions and generate the trajectories for various phases of the seismic refraction signals. The variance fractal dimension values, smaller for the refraction signal (/spl sim/1.25) than for noise (/spl sim/1.55), represent an important transition feature at the boundary between the signal and noise and provide us with a very useful application tool for separating the seismic refraction signals from noise.