Ultrasonic echoes from small or subtle defects in metals may be masked by competing “noise” echoes which arise from the scattering of sound by grains or other microstructural elements. Algorithms for estimating the detectability of such defects consequently require quantitative models for microstructural noise. In previous work [1,2] we introduced an approximate noise model for normal-incidence immersion inspections using tone-burst pulses, and we used the model to estimate signal/noise ratios for brittle (hard-alpha) inclusions in titanium alloys. In the present work we consider an extension of that noise model to inspections using broadband incident pulses. Like its predecessor, the broadband noise model neglects multiple scattering events, and applies to low-noise, low-attenuation materials. The broadband model provides an expression for the root-mean-square (rms) average amplitude of a given spectral component of the noise, computed on a finite time interval greater than the duration of the pulse. The model can be used to analyze backscattered noise to extract a Figure-of-Merit (FOM) for noise severity which is a property of the specimen and is independent of the measurement system. Conversely, if the FOM of the specimen is known, the model can be used to predict average noise spectral characteristics and average noise levels for various inspection scenarios.