To predict a retention time of analyte is already thoroughly researched topic. The knowledge of retention time value could serve many purposes, in LC-MS driven proteomics to know retention time of peptide may help to deal with acquired mass spectrometric data in order to maximise the information gain by means of number of identified peptides. Among the two principal approaches how to determine the retention time of compound, hard and soft modelling, the soft models have the advantage of being based on black box principle, what means the relations between respective retention time and compound properties in combination with separation system properties can be found without any exact knowledge of physico-chemical relation between them. The proper choice of compound descriptors and relevant settings of separation system may result in precise determination of retention time. As for the proteomics application, once it is possible to make a link between tryptic peptide sequence and its retention time, more justified identifications can be made out of collected data. The review deals with summarisation of contemporary knowledge of soft model based prediction of peptide retention time with a special attention given to artificial neural networks.