Extreme sub-daily precipitation is a major driver of damaging hazards such as flash and urban floods, making accurate rainfall modeling essential for infrastructure design and risk mitigation. Statistical methods for synthesizing rainfall across timescales, including temporal disaggregation and statistical downscaling, depend strongly on the underlying marginal distribution of rainfall intensities. While daily rainfall distributions have been extensively studied, the suitability of probability distributions for sub-daily rainfall across continuous aggregation timescales remains unexplored. This study evaluates three candidate distributions (Burr XII, Generalized Gamma, and Exponentiated Weibull) for representing sub-daily rainfall over timescales ranging from one hour to ten days. Using the quality-controlled Global Sub-Daily Rainfall dataset, which covers diverse climate zones and weather conditions worldwide, we examine the probability of dry periods and sample L-moment ratios (τ₂–τ₅) at each timescale. We explore whether the candidate distributions can reproduce the observed L-moment ratios before estimating their parameters. Two L-moment parameter estimation approaches are compared: one based on the lower-order pair (τ₂, τ₃) and another using the higher-order pair (τ₂, τ₅).None of the distributions represents all stations across all timescales. The Exponentiated Weibull is the most consistently feasible, while the Generalized Gamma performs poorly at short timescales but improves with aggregation. Burr XII shows less consistent performance and generally declines at longer timescales. The (τ₂, τ₅) criterion substantially improves feasibility, particularly for Burr XII and Generalized Gamma at fine scales, and produces lower quantile-fit errors than (τ₂, τ₃). Overall, checking distributional feasibility and accounting for rainfall-tail behavior are more important than the choice among these distributions.