Advanced infrared sounders will provide thousands of radiance data at every observation location. The issue of reducing the number of data to be used has been addressed in the context of optimal linear estimation theory, using simulated IASI data computed from a set of about 500 representative atmospheric profiles, with observation errors taken from current estimated values and background errors typical of a state-of-the-art Numerical Weather Prediction (NWP) system. Several methods have been tried to select a set of the most useful channels for each individual atmospheric profile. These are two methods based on the Data Resolution Matrix (DRM) for which one picks up channels corresponding to the largest diagonal elements of the DRM, one method based on the Jacobian matrix selecting the channels with the most appropriate weighting functions for retrieving each parameter at each level, and one iterative method selecting sequentially the channels with the largest information content and updating the error covariance matrix of the current analysis according to this new information. The methods based on the DRM were found to be more relevant for a posteriori diagnostics than for prior channel selection, whereas the Jacobian method and the iterative method were suitable for the problem. The iterative method was found to always produce the best results, but at a much larger cost than the Jacobian method. On the whole profile database, one variant of the iterative method has also been tried in an attempt to reduce its cost. It is an iterative method for which a "constant" channel selection is applied to each profile, this mean channel selection having been computed beforehand on the database. Results show that this "constant" iterative method is very promising, with results of intermediate quality between the ones obtained for the optimal iterative method and the Jacobian method. The practical advantage of this method for operational purposes is that the same set can be used for various atmospheric profiles.