Abstract: This study evaluates the forecast performance and error sources for precipitation in Guizhou of three Sub-seasonal to Seasonal (S2S) models—the China Meteorological Administration (CMA), the Institute of Atmospheric Physics, Chinese Academy of Sciences (IAP-CAS), and the U.S. National Centers for Environmental Prediction (NCEP)—during the flood seasons of 2021-2023, and conducts error correction experiments for the CMA model. Results show that NCEP performs best, with its threat score (TS) and percentage correct (PC) significantly higher than the other two models; IAP-CAS ranks second, while CMA exhibits the largest errors. All three models share common characteristics: higher forecast skill in the early flood season than in the late flood season, and larger errors over northeastern and southeastern Guizhou. Regarding error sources, CMA presents a unique meridional dipole structure in 500-hPa geopotential height (underestimation over subtropical regions and overestimation over mid-high latitudes), with its precipitation underestimation mainly attributed to systematic weakening of the East Asia-Pacific (EAP) teleconnection pattern gradient. IAP-CAS shows nearly full-domain underestimation, with precipitation underestimation dominated by insufficient dynamic forcing of the upper-level jet and shifts in the lower-level subtropical high pattern. NCEP, despite exhibiting similar circulation errors to the other models, shows opposite precipitation overestimation, suggesting that its errors may be more related to model physical parameterization. Based on the understanding of the CMA"s error characteristics, this study designs two nonparametric percentile mapping correction schemes: direct correction (CM1) and Madden–Julian Oscillation (MJO)-constrained correction (CM2). Comparative verification shows that CM1 effectively improves CMA"s forecast skill for 5–20 mm precipitation, while CM2 achieves further improvement by incorporating MJO constraints. The results provide a correction basis for sub-seasonal precipitation forecasts in Guizhou and offer a scientific reference for the interpretative application of S2S models.