Magma viscosity is one of the most critical physical properties controlling magma transport dynamics and eruptive style. Magma viscosity strongly depends on the melt phase composition (including dissolved volatile phases) and temperature, and, subordinately, on the crystal and bubble cargo. Several studies have experimentally investigated the dependence of melt viscosity on composition and temperature. However, recent studies have demonstrated that volcanic melts can be prone to nanocrystallization and dehydration during viscosity measurements. Such phenomena affect the reliability of experimental data and jeopardize the predictive ability of previous empirical models of magma viscosity. Here, we demonstrate the magnitude of inaccuracies in the determination of melt viscosity by presenting a new viscosity model of Stromboli basalt that considers the water-dependence of the glass transition temperature (Tg), measured via differential scanning calorimetry (DSC), and the melt fragility (m) derived by Brillouin spectroscopy. While anhydrous Stromboli basalt is not prone to nanocrystallization, we show that Fe-Ti-oxides are rapidly formed in the hydrous melt during viscosity measurements. Compared to our parameterization, previous empirical models overestimated 2-5 times the melt viscosity at eruptive conditions. These differences can strongly affect the ability to predict magma dynamics and emplacement processes, which are ultimately the basis for risk assessment and decision-making during volcanic crises.