From a research sample of 138 corpses, divided into four subgroups of ambient storage temperature (0-5 degrees C, 6-10 degrees C, 11-15 degrees C and 16-23 degrees C) four linear regression formulae of actual versus estimated post-mortem interval were obtained ('interval' formulae) using a single outer ear temperature measurement on both sides. This method showed the best correlation coefficient among five other methods previously proposed for time of death determination (rectal temperature, vitreous K+, CSF K+, blood log NA+/K+ and log Cl-), however its results were less accurate than those obtained with a multivariate equation combining several of the above mentioned methods. Eventually an equation expressing time of death (TOD) as a function of outer ear temperature (OE T degrees) and ambient temperature was also established from the whole research sample ('global' formulae). On a different sample of 141 corpses the regression formulae ('interval' and 'global') for the outer ear temperature were compared to three methods based on a single rectal temperature measurement ('rule of thumb' 1 and 2, Henssge nomogram) and therefore useful at the scene; the results of all methods were compared within the four subgroups of ambient temperature as well as in three subgroups of different post-mortem interval lengths (< 7 h, < 10 h, < 15 h). In all cases the outer ear temperature formulae provided better results than the rectal temperature methods (especially Henssge nomogram and rule of thumb 1). Moreover they did not show any post-mortem plateau which was present in almost 30% of cases when rectal temperature was measured in corpses kept at ambient temperature above 15 degrees C. Our results show that outer ear temperature measurement is the method which provides the best simplicity/quality ratio and should therefore be proposed for use at the scene when conditions are similar to those of our experiment (within buildings). A software equipped thermometer is required in order to use in each case the appropriate formula and confidence interval.