The reliability of the estimation of the size of scattering structures is assessed by realistic simulations and phantom experiments. The acoustic tissue model used in the simulation studies comprised a constant sound speed, homogeneous attenuation, and isotropic scattering. The scattering models considered were a discrete (spherical) model and two inhomogeneous-continuum models. The latter were characterized by an exponential and a Gaussian autocorrelation function, respectively. The backscattering spectra were, over the range from 5 to 10 MHz, fitted to linear, power, and autocorrelation functions of the three scattering models. The effects of the fitting function, the attenuation-either in an intervening layer or within the region of interest (ROI)-of the transmission pulse, and a spread in the scatterer sizes on the accuracy and the precision of the size estimates were assessed. The attenuation in the intervening tissue layer(s) as well as in the ROI itself has a significant effect on the accuracy of the size estimates and needs to be corrected. When performing the attenuation correction the inaccuracy of the attenuation estimate of the intervening layer leads to a large bias in the estimated scatterer size. Experimental results support the conclusion that scatterer size is a feasible tissue characterization parameter.