Iron mass balances for eight lakes (76 lake-years of data) on the Canadian Precambrian Shield were measured including catchment, atmospheric and internal sources of iron. These data were compared to or combined with available data from the literature on iron fluxes in other North American and European Lakes. In most of these lakes, internal iron load was important; it resulted from anoxic sediment release since the lakes' hypolimnia became partly anoxic during summer stratification. Iron retention, based on external iron load only and determined from the ratio of incoming minus outgoing iron mass and incoming mass (Rext), was correlated to internal load and not to morphometry and hydrology of the lake. On the other hand, when internal load was included in the calculated retention, (Rtot), retention was correlated with morphometry and hydrology, and not internal load. An average annual settling velocity of iron, computed from the relationship of Rtot with the annual water load (qs), ranged from 12 to 26 m yr-1 depending on the data set. A simple mass balance model was used to predict annual average iron concentration. Although model predictions and observed concentrations were significantly correlated, annual average concentrations were underestimated by the model in the Precambrian Shield lakes, but overestimated by atmospherically loaded and acid-stressed lakes from the Sudbury, Ontario region. Inclusion of a term related to organic acid content rendered the mass balance equation more general. Furthermore, the inclusion of dissolved organic acid concentration, colour of the lake water and loss on ignition of the surface sediment proved to be important secondary variables.