Strategies for improving the speed and performance of pattern recognition in the tracking detectors of hadron collider experiments are presented. They are based on fast algorithms which aim at preselecting the hits from the underlying physics event, while filtering out hits from noise and pile up. Quantitative results in terms of timing and efficiency are presented in the context of the ATLAS experiment at the LHC; an extrapolation to other hadron collider conditions, such as the Tevatron, is also discussed.