Cardiac resynchronization therapy (CRT) is an established nonpharmacological treatment for patients with heart failure (HF), reduced left ventricular (LV) ejection fraction, and a wide QRS complex. Although the therapy was developed 30 years ago and approved by the Food and Drug Administration in 2001, attempts to improve it have never stopped. Such improvements have been facilitated by combining knowledge from bench (basic science), bits (computer modeling), and bedside (clinical studies); these issues are addressed in the present review. Improvements include better patient selection, positioning of the LV lead, pacing from multiple sites, and optimization of atrioventricular and ventriculo−ventricular intervals. Overall, patterns of electrocardiographic and echocardiographic (strain) signals appear to be more useful for patient selection than timing intervals (QRS duration, time-to-peak shortening). Quadripolar leads have significantly improved CRT outcome due to increased electrical and mechanical lead performance (avoiding phrenic nerve stimulation and improving lead stability), but also thanks to the flexibility offered by the novel leads to avoid in-scar pacing. The benefit of multiple site stimulation over optimal conventional biventricular pacing seems small and is awaiting evidence from large trials. There is rapidly growing interest in merging imaging information to guide positioning of the LV lead in late activated regions without scar and in LV lead positions other than the epicardial coronary veins (LV endocardium, His bundle, LV septum). All these developments look promising but await further clinical validation. Finally, computer modeling is rapidly becoming important in understanding the substrate for CRT, in improving and assisting patient selection, as well as in guiding therapy planning.