Ultrasonic pulse echo techniques have proven particularly valuable in locating discrete flaws, such as cracks or inclusions, in metal components. It has long been recognized that many factors influence the echo amplitude from a defect. Three factors of particular significance are the angle of incidence between pulse and defect, the shape and size of the defect, and the roughness of its surface. A diffraction model is developed which includes these parameters and successfully accounts for the experimentally observed reflections of ultrasonic pulses from a variety of surfaces. The equations describing reflection in the frequency domain essentially predict the complex frequency spectra of pulses reflected from surfaces of various geometries: by Fourier transforming them, expressions describing the actual pulse waveforms reflected from the surfaces have been obtained. The predicted effects of surface roughness, size, shape, and defect orientation on both the reflected pulse waveform and its frequency spectrum are confirmed by experiment. The converse problem of using the reflected pulse to infer information about the reflecting surface via a deconvolution technique is also demonstrated.