Systematically determining the discriminatory power of various rainfall properties and their combinations in identifying debris flow occurrence is crucial for early warning systems. In this study, we evaluated the discriminatory power of different univariate and multivariate rainfall threshold models in identifying triggering conditions of debris flow in the Jiangjia Gully, Yunnan Province, China. The univariate models used single rainfall properties as indicators, including total rainfall (Rtot), rainfall duration (D), mean intensity (Imean), absolute energy (Eabs), storm kinetic energy (Es), antecedent rainfall (Ra), and maximum rainfall intensity over various durations (Imax_dur). The evaluation reveals that the Imax_dur and Eabs models have the best performance, followed by the Es, Rtot, and Imean models, while the D and Ra models have poor performances. Specifically, the Imax_dur model has the highest performance metrics at a 40-min duration. We used logistic regression to combine at least two rainfall properties to establish multivariate threshold models. The results show that adding D or Ra to the models dominated by Eabs, Es, Rtot, or Imean generally improve their performances, specifically when D is combined with Imean or when Ra is combined with Eabs or Es. Including Ra in the Imax_dur model, it performs better than the univariate Imax_dur model. A power-law relationship between Imax_dur and Ra or between Eabs and Ra has better performance than the traditional Imean-D model, while the performance of the Es-Ra model is moderate. Our evaluation reemphasizes the important role of the maximum intensity over short durations in debris flow occurrence. It also highlights the importance of systematically investigating the role of Ra in establishing rainfall thresholds for triggering debris flow. Given the regional variations in rainfall patterns worldwide, it is necessary to evaluate the findings of this study across diverse watersheds.