CMS has a two level trigger system. The first stage is hardware based and provides fast trigger decisions up to a rate 100 kHz. The second stage, known as the High-Level Trigger, is entirely software based and required to provide a trigger decision within 40 ms and a rejection factor of a thousand to achieve a write-to-disk rate of 100 Hz. One of the most CPU-intensive tasks within the High-Level Trigger is the reconstruction of tracking hits using raw data from the strip tracker. This study profiles the performance of these reconstruction algorithms. Even at low luminosities, the average processing time is 5.5 s, which already exceeds the HLT budget. A new schema, optimised for speed and performance, has been developed to reconstruct hits within regions-of-interest only. For the entire sub-detector, hit reconstruction times are reduced to 140 ms. Since only 10 % of High-Level Trigger events are expected to require track reconstruction, the average contribution per event is then ∼14 ms i.e. 30 % of the full budget. Regional reconstruction is tested over Z0 → e+e− events, by unpacking in η-ϕ windows of 0.16 × 0.16 around seeds identified in the calorimeter. In this case, only 2 ± 1 % of the silicon strip tracker raw data is reconstructed in 5 ± 3 ms (or an average contribution per event of 0.5 ms) whilst maintaining 99 % of the original dielectron trigger efficiency.