The Trigger System of the ATLAS experiment reduces the rate of events produced by proton-proton collisions at CERN's Large Hadron Collider (LHC) in three successive steps from 40 MHz to ~ 100, 1 and 0.2 kHz respectively. The ATLAS Second Level Trigger is original in several ways. It makes use of information provided by the First Level Trigger which identifies Regions of Interest (RoI) indicating where the most significant activity has occurred within the detector. Accessing detector data in RoIs only reduces the estimated 100 Gbytes/s data rate by a factor 100. Appart from a custom interface to acquire the RoI information, the Second Level Trigger is implemented in software. Another cost saving approach is the development of Trigger Selection software in an offline environment using a common framework for the High Level Trigger and Reconstruction Software. Consequently, the Second Level Trigger draws on software developed in two largely independant domains: real time oriented dataflow software combined with offline selection software. In this paper we report on experience gained and results obtained with second generation prototype software of both domains. Test of the performance of the data collection have been carried out on testbeds consisting of PCs running Linux and interconnected by Gbit Ethernet switches. The selection software has been tested using simulated detector data preloaded in detector readout buffers.