A generic toolbox for data assimilation called COSTA (COmmon Set of Tools for the Assimilation of data) makes it possible to simplify the application of data assimilation to models and to try out various methods for a particular model. Concepts of object oriented programming are used to define building blocks for data assimilation systems that can be exchanged and reused. The main building blocks in COSTA are the model component and the stochastic observer component. These components can be created by wrapping existing code. The LOTOS-EUROS air quality model will be used for operational smog and aerosol forecasts in the Netherlands in the near future. The COSTA framework will be used in this operational environment to implement the data assimilation techniques. As a first steps towards this operational system the model component of LOTOS-EUROS is created and the performance of various Kalman filter based data assimilation techniques are compared for a real live case study. The ensemble Kalman filter and the ensemble square-root filters converged faster than the other tested techniques for the selected case study setup. © 2009 Elsevier Ltd. All rights reserved.