Abstract: Based on the 24 h precipitation ensemble forecast products from three global numerical forecast centers (CMA, ECMWF, and NCEP) and surface precipitation observations in the Liaoning province from April 1 to June 30 from 2016 to 2018 and using Threaten Score, Bias, Talagrand and Brier Score methods, the forecast errors for spring soaking precipitation of the above-mentioned products in corresponding time were compared and analyzed.The results show that the dispersions of the ensemble forecast systems of the three centers are all small and their Talagrand diagrams all have a U-shaped distribution.Specifically, each ensemble forecast system overestimates the small magnitude of precipitation and has a high false alarm, and is incapable in forecasting the large magnitude of precipitation and underestimates the extreme value, which is easy to produce forecast deviation.In addition, by comparing the deterministic and probabilistic test results, it is found that TS value is higher and B values are closer to 1 for ECMWF relative to the other two forecast centers, implying that ECMWF has less false alarms than the other centers for spring soaking precipitation forecast in Liaoning province.Besides, the comparisons on the BS value and its decomposition score value also show that ECMWF is better than the other two centers.In summary, with the spring precipitation forecasts closest to the observations in Liaoning province and the best test results, ECMWF is available for forecast service work in the future as the main reference.