Abstract: In order to improve the relative humidity analysis and forecast of the CMA-MESO (China Meteorological Administration Mesoscale model) 3DVar (3-Dimensional Variational) data assimilation system at 3 km resolution, a new observation operator for 2 m relative humidity observations and the tangent linear as well as its adjoint have been coded, tested, and then introduced into the CMA-MESO 3DVar operational assimilation system. The T639 (T639L60 Global Medium Range Forecast System, 0.28125°×0.28125°) analysis of results for three months (1 June to 31 August, 2015) shows that the difference in relative humidity between 2 m and the lowest model level is a function of the Richardson number. Therefore, 2 m relative humidity can be directly compared to relative humidity at the lowest model level. The observation operator assumes that relative humidity is constant in unstable surface layer. The results show that more 2 m relative humidity observations can be used in the 3DVar assimilation system with this scheme, and the relative humidity analysis is more close to observations. The EST (Equitable Threat Score) value of summer (1—31 July, 2018) experiments is higher when 2 m relative humidity observations are assimilated using this scheme. The assimilation of surface relative humidity observations seems to be improved.