Cold thermoregulatory models (CTM) have primarily been developed to predict core temperature (T(core)) responses during sedentary immersion. Few studies have examined their efficacy to predict T(core) during exercise cold exposure. The purpose of this study was to compare observed T(core) responses during exercise in cold water with the predicted T(core) from a three-cylinder (3-CTM) and a six-cylinder (6-CTM) model, adjusted to include heat production from exercise. A matrix of two metabolic rates (0.44 and 0.88 m/s walking), two water temperatures (10 and 15 degrees C), and two immersion depths (chest and waist) were used to elicit different rates of T(core) changes. Root mean square deviation (RMSD) and nonparametric Bland-Altman tests were used to test for acceptable model predictions. Using the RMSD criterion, the 3-CTM did not fit the observed data in any trial, whereas the 6-CTM fit the data (RMSD less than standard deviation) in four of eight trials. In general, the 3-CTM predicted a rapid decline in core temperature followed by a plateau. For the 6-CTM, the predicted T(core) appeared relatively tight during the early part of immersion, but was much lower during the latter portions of immersion, accounting for the nonagreement between RMSD and SD values. The 6-CTM was rerun with no adjustment for exercise metabolism, and core temperature and heat loss predictions were tighter. In summary, this study demonstrated that both thermoregulatory models designed for sedentary cold exposure, currently, cannot be extended for use during partial immersion exercise in cold water. Algorithms need to be developed to better predict heat loss during exercise in cold water.