Abstract: Numerical weather prediction has evolved into a very complex system, hence effective and coordinated diagnosis of such a complex system has become a necessity.Complying with the principle of “based on the problem, understand the problem and then coordinate for the problem”, diagnosis for numerical weather prediction aims to solve problems and then to promote the development of numerical weather prediction system.Diagnosis has always been an active area of research and development at the European Centre for Medium-Range Weather Forecasts (ECMWF).A diagnostic strategy is essential for ECMWF to face “Where to go from here”.Diagnosis involves all aspects and components of the numerical weather prediction system.Data assimilation is an integral component of numerical weather prediction.It is the “coal-face” where forecasting model and observations“collide” and therefore the scope of diagnosis for data assimilation is immense.Centering on CMA four-dimensional variational assimilation system, this paper investigates the diagnosis work of advanced meteorological centers in the world, and then presents the top-level design of CEMC’s data assimilation diagnosis system in face of its fundamental and necessary components.Based on the program flow of the assimilation system and the internal logic of developing an operational assimilation system, its architecture includes diagnostics of input information, optimization process, output result, pre-specified parameters and access to new types of observation data, so as to initially form the assimilation diagnosis ability on a regular routine basis.Finally, the author puts forward the main priority tasks and initial thinking to start this action plan, especially highlighting its front-end basis and supporting conditions, e.g.generalized, shared and highly flexible database.The paper also discusses the requirement for a consensus on this work based on problem-oriented onset, scientific understanding and collaboration.