In ultrasonic inspections for small or subtle defects in metals, defect signals may be obscured by grain noise echoes which arise from the scattering of sound by the microstructure of the metal. Models for predicting microstructural noise levels are consequently essential for accurately assessing the reliability of the ultrasonic inspections. Existing noise models, like the independent scatterer model (ISM) [1], are capable of predicting only average noise characteristics, such as the root-mean-square (rms) noise level. Average noise levels, although useful, are not sufficient for assessing detection reliability. One needs to know the manner in which noise signals are distributed about their average level. The expected peak noise level, for example, effects the rate of “false calls”, in which noise signals are mistaken for echoes from critical defects. In this work, we present a Monte-Carlo method for simulating time-domain noise signals observed in pulse/echo immersion inspections of metal components. The method predicts simulated time-domain noise signals, and hence can be used to determine both average and peak noise levels. We assume that the backscattered noise is dominated by the single- scattering of the incident beam by individual metal grains. The metal volume is represented as an ensemble of spherical, single-crystal grains whose centers and orientations are randomly chosen. Grain radii are determined by the nearest-neighbor distances and volume conservation. The backscattered voltage signal from each grain is calculated by treating the grain as an anisotropic scatterer in the homogeneous average medium formed by the other grains. Backscattered signals from all grains are summed to determine the total noise signal.