Analysis errors are a desirable important byproduct of data assimilation systems (DAS). They provide both a measure of accuracy of each analysis produced by the DAS and the necessary initial condition for error prediction accounted for in procedures such as the parameterized Kalman filter. The method proposed by Riishojgaard (2000; Q. J. R. Meteorol. Soc., 126, 1367-1385) to calculate analysis errors in the context of the physical-space statistical analysis system (PSAS) has been implemented in the DAO DAS. However, its storage and computational requirements are quite overwhelming in practice. The present work aims at constructing an acceptable configuration for this implementation that can be used operationally. Particularly, we are investigating the advantages of using the so-called reduced-PSAS (Lyster and Guo 2001, personal communication) as a viable possibility to cut down the computational cost of calculating analysis errors. In the reduced-PSAS, the error covariance matrices are approximated locally to reduce the total number of floating point operations. Furthermore, algorithmic strategies making use of the Message-Passing Interface are being considered as well. Results from this work will be presented and discussed during the conference.