Biodegradable naturally occurring adsorbents derived from waste precursors are essential for water sustainability. This study investigates using modified cellulose nanostructure (m-CNS) with thiols from wood pulp as a waste source to remove Cr(VI) ions from aqueous solution under different conditions, such as temperature, initial dye concentration, and contact time. The equilibrium adsorption of Cr(VI) is assessed at various temperatures (30, 40, and 50 °C) and concentrations (10, 20, 30, 40, and 50 mg L⁻¹). The m-CNS is detected by Fourier transform infrared spectroscopy (FT-IR), scanning electron microscopy (SEM), and dispersive X-ray spectroscopy (EDS). Experiments are being carried out to investigate the removal of Cr(VI) ions in equilibrium state. The results showed that the highest percent removal of Cr(VI) ions was 95.95% at pH = 4.0 and after a relatively short adsorption time (80 min). The experimental data is presented using a diverse range of seven isotherm models. There are four models with two parameters: Freundlich, Langmuir, Dubinin-Radushkevich, and Temkin. In addition, three models with three parameters, namely the Redlich-Peterson, Sips, and Toth models, are employed to analyze the experimental adsorption data comprehensively. The depth of our analysis is further enriched using six error functions: the chi-square test (χχ2), the sum of squares of the errors (SSE), the derivative of Marquard’s percent standard deviation (MPSD), the average relative error (ARE), the sum of absolute errors (EABS), and the coefficient of determination. Unlike the Dubinin-Radushkevich isotherm, linear and non-linear regression procedures produced equivalent results for two-parameter isotherms at different temperatures. This is especially noteworthy since the Freundlich, Langmuir, and Temkin isotherms, which provided the greatest fit to the data, are frequently utilized in isotherm modeling and adsorption research. Three-parameter isotherms yielded conflicting linear and non-linear model findings across different temperatures. Furthermore, the findings show that the most optimum error function for prediction was χχ2.