Most mass spectrometry based metabolomics studies are semiquantitative and depend on efficient normalization techniques to suppress systematic error. A common approach is to include isotope-labeled internal standards (ISs) and then express the estimated metabolite abundances relative to the IS. Because of problems such as insufficient chromatographic resolution, however, the analytes may directly influence estimates of the IS, a phenomenon known as cross-contribution (CC). Normalization using ISs that suffer from CC effects will cause significant loss of information if the interfering analytes are associated with the studied factors. We present a novel normalization algorithm, which compensates for systematic CC effects that can be traced back to a linear association with the experimental design. The proposed method was found to be superior at purifying the signal of interest compared to current normalization methods when applied to two biological data sets and a multicomponent dilution mixture. Our method is applicable to data from randomized and designed experiments that use ISs to monitor the systematic error.