AbstractA major revision is described to the analysis component of the Meteorological Office data assimilation scheme for both the global and the regional numerical weather prediction models. Initialization by repeated insertion of data and divergence damping is retained but a more flexible modified successive‐correction scheme replaces optimum interpolation for the analysis step. The scheme is derived as an approximate iterative solution to a general equation for the optimal analysis. Within iterations, the analyses of each model variable are performed sequentially and multivariate increment fields are derived for dynamical balance at intermediate stages. The horizontal influence area of data is substantially increased, giving a beneficial impact in data‐sparse regions. All data contribute to the analysis at each gridpoint within their radius of influence. The observation weights are recalculated every model time‐step, allowing asynoptic data to be used at their true valid time, though this feature is found to have only slight impact on forecasts. Each analysis is factorized into a vertical step, then into a horizontal step. The structure of each is described in detail, with system parameters listed in appendices.Results of two kinds are presented. Sensitivity experiments highlight the effects of some individual constituents of the scheme; operational trials against the old scheme reveal the impact of the whole system. The new analyses are better balanced and lead to a better short‐period forecast, especially for the wind field. Medium‐range impact is clearly beneficial in the southern hemisphere but ambiguous in the northern hemisphere. Precipitation forecasts in the regional model are