Parameter estimation is crucial to the accuracy of Neugebauer model in characterizing digital halftone printers. In this paper, a novel total least square (TLS) regression technique is proposed for the model parameter estimation problem for dot-on-dot screen printers. Compared to the traditional least square (LS) approach, the TLS method is physically more appropriate because it accounts for errors in both the measured reflectance of the selected primaries and the modeled reflectance. At the same time, TLS provides increased mathematical flexibility in fitting the Neugebauer model to the experiment data. Both the TLS and LS techniques are tested on a Xerox color printer with a dot-on-dot halftone screen. Compared to the LS based techniques, the TLS methods yield a more accurate model for the printer.