We show that the standard analysis procedures as sum rule application and multipole-moment analysis for XMCD spectra can fail for magnetic samples of present interest. Two examples will be given: 1) The induced magnetic moments in ultrathin films of the light 3d elements Ti, V and Cr at the interface to Fe cannot be determined by the XMCD sum rule or multipole-moment analysis at the L2,3 edges. This is due to correlation effects which result in the deviation of the intensity ratio (branching ratio) from its statistical value. To address this point we established a double-pole approximation within time-dependent density functional theory. 2) The analysis of the L2,3 XMCD of rare earth elements is not only hampered by the appearance of electric quadrupolar contributions (E2) in addition to the dipolar contributions (E1). Even after separation of the two, one determines the wrong sign of the induced 5d moment by the sum rules. This originates from the spin dependence of the transition matrix elements. To tackle these difficulties we compare the experimental spectra to ab initio calculations of the entire isotropic XAS and the dichroic signal.