Three important technical innovations are reported here towards Raman-activated cell sorting. Firstly, a microfluidic chip made of quartz is introduced which integrates injection of single cells, trapping by laser fibres and sorting of cells. Secondly, a chip holder was designed to provide simple, accurate and stable adjustment of chips, microfluidic connections and the trapping laser fibres. The new setup enables to the collection of Raman spectra of single cells at 785 nm excitation with 10 s exposure time. Lastly, a new type of modelling the various background contributions is described, improving Raman-based cell identification by the classification algorithm linear discriminant analysis. Mean sensitivity and specificity determined by iterated 10-fold cross validation were 96 and 99 %, respectively, for the distinction of leucocytes extracted from blood, breast cancer cells BT-20 and MCF-7, and leukaemia cells OCI-AML3.