Broad-scale, long-term studies of water quality (WQ) are critical to understanding global-scale pressures on inland waters, yet they are rare. This data product, LAGOS-US LANDSAT, addresses this gap by providing remote sensing-derived WQ estimates from machine learning models trained on in situ data of six essential WQ variables for 136,977 lakes in the continental US from 1984-2020. The dataset includes: (a) 45,867,023 sets of whole-lake water reflectances for six individual bands and 15 band ratios; (b) 740,627 matchups with in situ data for lake WQ data for chlorophyll, Secchi depth, true color, dissolved organic carbon, total suspended solids, and turbidity; and, (c) predictions from each reflectance set for all six WQ variables across the 37 yr period. Variance explained for the predictions ranged from 20.7% for TSS to 63.7% for Secchi. Data extraction from individual scenes was quality-controlled based on cloud-cover and pixel quality, and we tested and validated key parts of the workflow to inform future water quality studies using the Landsat platform.