The application of predicted LC retention time to support metabolite identification was evaluated for a metabolomics MS/MS database containing 532 compounds representative for the major human metabolite classes. LC retention times could be measured for two C18 type columns using a mobile phase of pH = 3.0 for positive ESI mode (n = 337, 228) and pH = 8.0 for negative ESI mode (n = 410, 233). A QSRR modelling was applied with a small set of model compound selected based on the Kennard-Stone algorithm. The models were implemented in the R environment and can be applied to any library. The prediction model was built with two molecular descriptors, LogD2 and the molecular volume. A limited set of model compounds (LC CalMix, n = 16) could be validated on two different C18 reversed phase LC columns and with comparable prediction accuracy. The CalMix can be used to compensate for different LC systems. In addition, LC retention prediction was found, in combination with SWATH-MS, to be attractive to eliminate false positive identification as well as for ranking purpose different metabolite isomeric forms.