Global compilations of surface heat flow data from stable, Precambrian terrains show a statistically significant secular change from 41±11 mW/m 2 in Archean to 55±17 mW/m 2 in Proterozoic regions far removed from Archean cratons. Using the tectonothermal age of the continents coupled with average heat flow for different age provinces yields a mean continental surface heat flow between 47 and 49 mW/m 2 (depending on the average, non-orogenic heat flow assumed for Phanerozoic regions). Compositional models for bulk continental crust that produce this much or more heat flow (i.e., K 2O>2.3–2.4 wt%) are not consistent with these observations. More rigorous constraints on crust composition cannot be had from heat flow data until the relative contributions to surface heat flow from crust and mantle are better determined and the non-orogenic component of heat flow in the areally extensive Phanerozoic regions (35% of the continents) is determined. We calculate conductive geotherms for 41 mW/m 2 surface heat flow to place limits on the heat production of Archean mantle roots and to evaluate the significance of the pressure–temperature ( P– T) array for cratonic mantle xenoliths. Widely variable geotherms exist for this surface heat flow, depending on the values of crustal and lithospheric mantle heat production that are adopted. Using the average K content of cratonic peridotite xenoliths (0.15 wt% K 2O, assuming Th/U=3.9 and K/U=10,000 to give a heat production of 0.093 μW/m 3) and a range of reasonable crustal heat production values (i.e., ≥0.5 μW/m 3), we calculate geotherms that are so strongly curved they never intersect the mantle adiabat. Thus the average cratonic peridotite is not representative of the heat production of Archean mantle roots. Using our preferred estimate of heat production in the cratonic mantle (0.03 wt% K 2O, or 0.019 μW/m 3) we find that the only geotherms that pass through the xenolith P–T data array are those corresponding to crust having very low heat production (<0.9 wt% K 2O). If the lithospheric mantle heat production is higher than our preferred values, the continental crust must have correspondingly lower heat production (i.e., bulk crustal K, Th and U contents lower than that of average Archean granulite facies terrains), which we consider unlikely. If the xenolith P–T data reflect equilibration to a conductive geotherm, then Archean lithosphere is relatively thin (150–200 km, based on intersection of the P–T array with the mantle adiabat) and the primary reason for the lower surface heat flow in Archean regions is decreased crustal heat production, rather than the insulating effects of thick lithospheric roots. On the other hand, if the xenolith P–T points result from frozen-in mineral equilibria or reflect perturbed geotherms associated with magmatism, then the Archean crust can have higher heat producing element concentrations, lithospheric thickness can range to greater depths and the low surface heat flow in Archean cratons may be due to the insulating effects of thick lithospheric roots. An uppermost limit for Archean crustal heat production of 0.77 μW/m 3 is determined from the heat flow systematics.