Abstract: To improve the diagnostic scheme and the forecast results of precipitation types in numerical models, the surface observational and sounding data, precipitation type diagnosis products (PTYPE) from ERA5 reanalysis data, and temperature and dew point of different pressure levels and the surface are used in this study. The PTYPE diagnostic and observational sounding data for different precipitation types in the northern, southwestern, and Jiangnan area of China from the winter of 2022 to 2023 are compared to analyze the causes of forecast biases in the PTYPE products from the aspects of statistical conceptual models and physical processes, and then the improvement methods are proposed. The results are as follows. (1) The combination of dry and wet snow incidents recognized by PTYPE is consistent with the observed snow events in the northern and Jiangnan area. However, in the southwestern area, the PTYPE tends to overestimate the number of snowfall events, suggesting significantly lower reliability in this area. (2) The PTYPE diagnostic scheme is not consistent with the fundamental physical process of freezing rain formation, overlooking the droplets' supercooled characteristics, which could be the primary reason for the “false alarm” of the freezing rain in the PTYPE product. (3) For diagnosing ice pellets, the PTYPE scheme requires a 2-m surface temperature below 0 °C, which is usually not the case for observations. This could be the main reason for the missed alarm of ice pellets. Additionally, the diagnostic scheme only requires that the proportion of the solid particles near the surface is no less than 50%, rather than a percentage of most or even all, which leads to “false alarm”. (4) Base on comprehensive analysis, an improved PTYPE scheme is proposed: using the surface wet-bulb temperature as a criterion for sleet and snow; adding near-surface droplet temperature for freezing rain; removing the surface temperature in the ice pellet scheme, and appropriately lowering the threshold of "near-surface liquid water content", which are expected to improve the accuracy of these precipitation types.