Precise on-line and bulk-density independent monitoring of moisture content in processing grain into flour is important information to improve the milling process efficiency and to obtain a high quality final product. However, since feed moisture content is difficult to measure, it is nearly never available in the milling industry. This work addresses a reliable solution to reduce the deleterious effect of the mass flow rate on the estimation of grain sorghum moisture content, a method which is based on the high correlation between grain moisture content and two of its dielectric properties, i.e., relative permittivity epsilon ', and loss factor epsilon ''. Thus, the object of this paper was to analyse the effectiveness of three dielectric models for estimation of sorghum moisture content at excitation frequencies in the range from 75 kHz to 5 MHz, for moisture contents from 13 to 23% w.b., and bulk densities ranging from 575 to 818 kg m(-3). Goodness of fit between experimental and predicted values of moisture content was based on the following statistical parameters: coefficient of multiple determination, standard error of calibration, worst-case error, root mean squared error, mean absolute percentage error, mean bias, fit index, and reduced chi-squared. Residual plot analysis and the principle of parsimony were also considered in the selection of the most effective dielectric model. The model based on the measurement of epsilon ' and In(epsilon '') at 5 MHz gave the best predictions of sorghum moisture content for the conditions studied. (C) 2013 Elsevier B.V. All rights reserved.