Reconstructions of the North Atlantic Oscillation Index (NAOI) for the last centuries are mostly based on statistical models linking this index with proxy records. The underlying assumption is that the relationship between the NAO and the proxy records is stable in time and independent of time scale. This assumption might not be physically substantiated, since at time scales of centuries, other processes, such as solar variability, might disturb the link between the NAOI and temperature-sensitive indicators. The statistical approach for NAOI reconstructions was tested using a climate simulation with a climate model driven by the external forcing of the last 490 years, as a surrogate climate. Two kinds of indicators were tested, air temperature and precipitation. It was found that the NAOI reconstructions based on precipitation are more reliable than the reconstructions based on temperature. Furthermore, the choice of geographical box has a non-negligible influence on the reconstructed NAOI.