Cardiac resynchronization therapy (CRT) is an increasingly popular therapy for heart failure patients with intraventricular conduction delay, such as left bundle branch block. However, clinical trials with CRT have revealed that about 30% of patients may be classified as non-responders mainly due to a lack of predictive, mechanism-based objective criteria for patient selection. In addition, even responders exhibit variable degrees of benefit from the therapy, implying that it is not always performed in an optimal fashion. In order to predict and optimize the outcomes of CRT on individual patients, methods for generating anatomically detailed three-dimensional finite element models of failing human heart and estimating various patient-specific model parameter values from clinical measurements in heart failure patients are developed. The three-dimensional finite element model was generated from cardiac computed tomography images obtained at end-diastole. The fiber architecture was modeled by computing the eigenvectors from nodal fields of diffusion coefficients obtained from diffusion tensor magnetic resonance imaging (DTMRI). The endocardial activation maps, which serve as initial conditions in estimating the three-dimensional activation sequence of the myocardium, were generated by fitting endocardial activation time measurements from electroanatomic mapping. Lastly, a method for estimating left ventricular volumes from transthoracic echocardiograms was proposed. These methods provide efficient techniques to estimate patient-specific model parameters and therefore, will prove useful in patient-specific model development and ultimately, in predicting clinical outcomes of CRT on individual heart failure patients.