The concentration and size fractionation of major and trace elements (TEs) linked to organo-mineral colloids along the landscape gradient soil water–peat mire–humic lakes–river–large oligotrophic lake have been investigated within two watersheds located in the subarctic region (Karelia, Russia) in the summer base-flow period. A large volume of natural waters was filtered in the field using cascade filtration and ultrafiltration (UF) with a progressively decreasing pore size [100, 20, 10, 5, 0.8, 0.4, 0.22, 0.1, 0.046, 0.0066 (100 kDa), 0.0031 (10 kDa) and 0.0014 µm (1 kDa)] followed by multi-elemental ICP-MS and dissolved organic carbon (DOC) analysis. In situ dialysis with 10 and 1 kDa molecular weight cutoff membranes was also conducted. According to the TE distribution among different size fractions in the continuum soil (mire)–humic lake–river–large lake, the following groups of elements could be distinguished: (1) elements significantly bound to the colloidal (1 kDa—0.22 µm) fraction and decreasing the proportion of this fraction from the feeding lake and stream to the terminal oligotrophic lake (Fe, Ti and U); (2) colloidally dominated (>80 %) elements exhibiting similar size fractionation in all studied settings (Y, REEs, Zr, Hf and Th); (3) elements appreciably bound to colloids (20–40 %) in organic-rich soil and bog waters and decreasing their colloidal fraction to approximately 10 % in the oligotrophic lake (Mg, Ca, Sr, Rb and Mo); and (4) elements linked to the colloidal fraction (40–80 %) and not demonstrating any systematic variation between different landscape units (Ni, Co, Cu, Cd, Cr, Mn, Zn and Pb). The in situ dialysis technique produced a similar picture of the colloidal size fractionation with generally lower proportions of elements in the colloidal fraction compared to the UF. Thermodynamic modeling of trace element speciation using available codes demonstrated that complexation with dissolved organic matter and adsorption at the surface of ferric colloids can adequately model the observed colloidal speciation of divalent metals. However, the modeling could not describe the distribution of trivalent and tetravalent hydrolysates (TE3+,4+) among different size fractions and cannot reproduce the experimentally observed proportion of their colloidal forms. Therefore, coprecipitation with organo-ferric colloids should be considered to account for the partitioning of TE3+,4+ between truly dissolved (<1 kDa) and various colloidal fractions. It follows from the results of this study that autochthonous processes of organic matter fractionation, such as (1) transformation of initially allochthonous soil-derived colloids via photo- and biodegradation or (2) new organic ligand production by plankton and periphyton, cannot appreciably affect the distribution of TEs among various size fractions of colloids and particles along the landscape gradient from soil water to terminal lake. Thus, even crucial changes in the environmental conditions, such as the river discharge regime and relative lake versus bog coverage, surface water pH, pCO2 and DOC changes that are likely to occur according to the “extreme” scenario of the climate warming in the Arctic, should not affect the size fractionation and, therefore, the potential aquatic lability and biodisponibility of TEs present as organic (divalent transition metals, alkaline earth metals) and organo-mineral (Fe, Al, insoluble trivalent and tetravalent elements and U) colloids.