This study presents an empirical approach to optimize conductivities within a torso model given simultaneous epicardial and body surface potential recordings. The conductivities of the lungs, skeletal muscle and torso cavity were estimated within a forward model by minimizing the relative error between computed and reference torso potentials using a standard gradient-based approach. The sensitivity of this approach was evaluated over different levels of geometric error and signal noise, and the gradient of the cost function was determined using both 1) finite differences and 2) an adjoint method. All conductivities were accurately estimated (<10% difference in value) with up to 0.20 mV signal noise and all levels of electrode localization error (up to 2.56 cm) using a finite difference approach. While the adjoint approach was more computationally efficient, a finite difference approach was more stable across different signals and more robust to noise.