[en] This thesis is concerned with electron/jet discrimination and b-jet tagging for the ATLAS second level trigger. Electron/jet discrimination at the global second level trigger has been investigated with two different methods, combined linear cuts and Artificial Neural Networks. Data without and with pile up were represented by 11 selection variables from the calorimeter, the preshower and the tracking systems, transition radiation tracker and silicon strip detectors. For data without pile up a rejection of 300 is achieved for 95% electron efficiency for both methods. For data with pile up the neural net proved better with a jet rejection of 100 and electron efficiency of 70% whereas the linear cut efficiency drops below 60%. The second level trigger has access to full granularity data from the tracking detectors in regions where the first level trigger found high transverse momentum jets. The separation between b-quark jets and u-quark jets at the second level trigger has been studied with an algorithm for finding multiple tracks in the pixels and silicon strip detectors. The algorithm employs histogramming of Hough transformed straight lines for pattern recognition. This algorithm was used for discrimination between b-jet and u-jet data of the decay H → bb-bar for m_H = 100 GeV. A rejection of 30 u-jets for a b-tagging efficiency of 50% has been achieved using the transverse impact parameter d₀ for jets with ≥ 3 reconstructed tracks. (author)