We initiated a large-scale bidirectional selection experiment in a genetically heterogeneous rat population (N/NIH stock, n = 152) to develop lines of low response trainers (LRT) and high response trainers (HRT) as a contrasting animal model system. Maximal treadmill running distance [meters (m)] was tested before (DIST(1)) and after (DIST(2)) standardized aerobic treadmill training over an 8 wk period (3 exercise sessions per week). Response to training was calculated as the change in exercise capacity (ΔDIST = DIST(2) - DIST(1)). A within-family selection and rotational breeding paradigm between 10 families was practiced for both selected lines. For the founder population, exercise training produced a 140 ± 15 m gain in exercise capacity with interindividual variation ranging from -339 to +627 m. After 15 generations of selection (n = 3,114 rats), HRT rats improved 223 ± 20 m as a result of exercise training while exercise capacity declined -65 ± 15 m in LRT rats given the same absolute training environment. The narrow-sense heritability (h(2)) for ΔDIST was 0.10 ± 0.02. The LRT and HRT lines did not differ significantly for body weight or intrinsic (i.e., DIST(1)) exercise capacity. Using pedigree records the inbreeding coefficient increased at a rate of 1.7% per generation for HRT and 1.6% per generation for LRT, ∼30% slower than expected from random mating. Animal models developed from heterogeneous stock and enriched via selection, as presented here, often generate extreme values for traits of interest and may prove more useful than current models for uncovering genetic underpinnings.