At least six urinary metabolite ratios of caffeine have been proposed as probes for in vivo CYP1A2 activity and three for in vivo NAT2 activity. Claims for the frequency distribution of the activity of CYP1A2 based on these empirical ratios have varied from log-normal to trimodal. We have examined the validity of these nine ratios by developing computer simulations using values reported in the literature for the kinetic parameters of caffeine and its metabolites. The results show that the sensitivity of the ratios to confounding variables is, in some cases, greater than their sensitivity to the activity of the enzyme that they are intended to mark. The six CYP1A2 ratios did not exhibit the same pattern of dependency on confounding variables which, in turn, resulted in different shapes of population distributions for each ratio as enzyme activity was varied systematically. Although the dependency of the three NAT2 ratios on confounding variable was less marked, they also showed different patterns of dependency. The outcomes of the simulations were consistent with much of the experimental data on caffeine metabolite ratios. To support the findings from the simulations, simplified equations for each metabolite ratio were derived which emphasize the dominant determinants. With some of the CYP1A2 ratios urine flow was significant to the point where its variance and heterogeneity between populations could lead to spurious detection of polymorphism in CYP1A2 function. Also, if the variability of a dominant confounding factor was high and sensitivity of the ratio to intrinsic CYP1A2 activity was low, any polymorphism in the latter would be obscured. When a specific time interval was defined for urine collection, this time was shown to be a critical factor in the ability to discriminate bimodality in some of the ratios, when a marked polymorphism in enzyme activity was assumed. Those ratios which have shown no evidence for bimodality in CYP1A2 function in experimental studies are inherently more discriminant of such heterogeneity compared to those ratios which have been claimed to detect polymorphism of CYP1A2 from experimental data. While recommending a 'best buy' from amongst the caffeine urinary metabolite ratios, we favour plasma/saliva indices (caffeine half-life or paraxanthine/caffeine ratio in a spot sample).