<p>Accurate prediction and modeling of subcooled flow boiling critical heat flux (CHF) are essential for the safe operation of two-phase heat transfer systems across a wide range of engineering applications. A fundamental challenge lies in the empirical and often inconsistent selection of dimensionless numbers, which affects correlation performance, mechanistic interpretation, and the design of similarity experiments. In this study, a data-driven dimensional analysis approach is employed to overcome the non-uniqueness of traditional dimensional analysis. A comprehensive subcooled flow boiling CHF dataset comprising 4238 data points across 11 fluids is compiled; to the best of the authors’ knowledge, this is the largest multi-fluid dataset currently available. Using this method, a new key dimensionless number, denoted as <i>π</i>_CHF, is discovered, showing a strong negative correlation with the modified boiling number (<i>Bo</i>*). The core of <i>π</i>_CHF is formed by a newly designed Evaporation number (<i>Ev</i>), representing the proportion of surface tension energy overcome during bubble growth relative to the total absorbed energy, and the gas phase Peclet number (<i>Pe</i>_g), characterizing the relative importance of convective versus diffusive heat transfer within the gas phase. Based on <i>π</i>_CHF, a new predictive correlation is developed, achieving an overall mean absolute percentage error (MAPE) of 12.55% across the dataset, demonstrating superior accuracy and broad applicability compared to nine existing models. These findings provide a promising basis for further mechanistic studies, model development, and the design of fluid-to-fluid similarity experiments, particularly for advanced nuclear energy systems, aerospace thermal management, and electronics cooling.</p>