A high‐resolution demonstration Numerical Weather Prediction (NWP) nowcasting system was developed and run in real time at the Met Office for the prediction of precipitation on a domain covering southern England and Wales. The hourly cycling nowcasting system, known as NDP (Nowcasting Demonstration Project), combined a 3 km resolution 4D‐Variational data assimilation (4D‐Var) and a 1.5 km resolution version of the Met Office Unified Model (UM) to provide hourly NWP analyses and forecasts for a period of 0–6 h. Central to the NDP was the rapid updating cycling and its timely delivery of analyses and forecasts using the latest conventional and sub‐hourly novel observations. In this article the benefits of using 4D‐Var assimilation compared to First Guess at Appropriate Time (FGAT) 3D‐Variational assimilation (3D‐Var) on forecasts of precipitation are considered by comparing model forecasts with radar‐derived hourly surface accumulations for the period of June 2012, using an objective, scale‐dependent verification scheme. It is shown that 4D‐Var assimilation has a positive impact on precipitation forecasting skills compared to the corresponding 3D‐Var assimilation for the whole nowcast period [T + 0, T + 6]. The 4D‐Var assimilation system produced forecasts with greater spatial accuracy and longer lead times with acceptable skill. By comparing 3D‐Var FGAT using sub‐hourly observations with 3D‐Var FGAT using only the observations closest to the analysis time, the sub‐hourly observations are seen to be beneficial in the hourly cycling convective‐scale system in both 3D‐Var FGAT and 4D‐Var. The forecast clearly benefitted from the use of extra, higher time frequency observations in both 3D‐Var and 4D‐Var.