In this paper, the dependence of correlations between spectral accelerations at multiple periods on magnitude (M) and distance (R) has been investigated. For this purpose, a relatively large dataset of ground motion records (GMRs), containing 1551 records with a wide range of seismic characteristics, was selected. It is shown that the difference in the correlation coefficient is statistically meaningful when the general GMR dataset is divided into two subsets based on an arbitrary M or R. The observed difference is more meaningful in the case of magnitude when compared with distance. The general dataset of GMRs was then divided into four separate subsets based on optimum values of M and R, so that the four obtained subsets were given the greatest dissimilarity in terms of the correlation coefficients. The correlation coefficients between spectral accelerations at multiple periods were calculated in the case of the four subsets and compared with the available correlations in the literature. The conditional mean spectrum was also calculated by means of the conventional correlation coefficients, as well as by using the proposed M and R dependent correlation coefficients. The results show that, despite the commonly available findings in the literature, this dependence is significant and should not be neglected in the conditional spectra calculation process. Copyright © 2013 John Wiley & Sons, Ltd.