Changing lake water total organic carbon (TOC) concentrations are of concern for lake management because of corresponding effects on aquatic ecosystem functioning, drinking water resources and carbon cycling between land and sea. Understanding the importance of human activities on TOC changes requires knowledge of past concentrations; however, water-monitoring data are typically only available for the past few decades, if at all. Here, we present a universal model to infer past lake water TOC concentrations in northern lakes across Europe and North America that uses visible-near-infrared (VNIR) spectroscopy on lake sediments. In the orthogonal partial least-squares model, VNIR spectra of surface-sediment samples are calibrated against corresponding surface water TOC concentrations (0.5-41 mg L-1) from 345 Arctic to northern temperate lakes in Canada, Greenland, Sweden and Finland. Internal model-cross-validation resulted in a R2 of 0.57 and a prediction error of 4.4 mg TOC L-1. First applications to lakes in southern Ontario and Scotland, which are outside of the model's geographic range, show the model accurately captures monitoring trends, and suggests that TOC dynamics during the 20th century at these sites were primarily driven by changes in atmospheric deposition. Our results demonstrate that the lake water TOC model has multiregional applications and is not biased by postdepositional diagenesis, allowing the identification of past TOC variations in northern lakes of Europe and North America over time scales of decades to millennia.