Past studies have demonstrated the improved diagnostic utility of the electrocardiogram by utilizing alternative recording sites to that used in the standard 12-lead. The present study proposes new lead sets based on the developments of innovative computational models. Utilizing 744 117-lead body surface potential maps the diagnostic value of different recording site combinations was evaluated. Three lead sets were derived; sites discriminating between subjects with evidence of old myocardial infarction and healthy subjects; sites discriminating between subjects with left ventricular hypertrophy and healthy subjects; and sites discriminating between the two disease types. A wrapper approach incorporating a nearest neigbour classifier was adopted to facilitate the lead selection process. Following 10-fold cross validation the recording sites chosen for LVH vs Normal, MI vs Normal and MI vs LVH yielded sensitivities of 89%, 89% and 75%, and specificities of 91%, 94% and 65%, respectively.