We compile upper mantle Pn velocities from seismic refraction/wide-angle reflection surveys in the southern Superior Province of the Canadian Shield and compare them with temperatures at the Moho deduced from heat flow data. Calculated Moho temperatures and Pn velocities correlate well, showing that in this area, Pn depends primarily on temperature. The obtained values of ∂V(Pn)/∂T depend weakly on the assumed value of Moho heat flow and are on the order of -6.0 ¿ 10-4 ¿ 10% km s-1 K-1, within the range of temperature derivatives obtained in laboratory studies of ultramafic rocks. Comparison between observed Pn velocities and predicted values for several mineralogical models at Moho temperatures allows constraints on both the Moho heat flow and the shallow mantle composition. For all Moho heat flows, undepleted (clinopyroxene-rich) mantle compositions do not allow a good fit to the data. For depleted mantle compositions, temperatures consistent with the observed Pn velocities correspond to values of Moho heat flow larger than 12 mW m-2. For our preferred Moho heat flow of 15 mW m-2, the best fit mantle composition is slightly less depleted than models for average Archean subcontinental lithospheric mantle. This may be due to rejuvenation by melt-related metasomatism during the Keweenawan rifting event. The similarity in Pn - T conversion factors estimated from this empirical large-scale geophysical study and those from laboratory data provides confidence in the absolute temperature values deduced from heat flow measurements and seismic studies.