The present study addresses the problem of respira- tory rate estimation from ECG-derived respiration (EDR) signals during paroxysmal atrial fibrillation (AF). Novel signal-to-noise ratios between various components of the ECG including the influence of respiration, measured by QRS ensemble variance, the amplitude of fibrillatory waves (f-waves), and the QRS amplitude are introduced to characterize EDR performance. Using an improved ECG simulation model accounting for morphological variation induced by respiration, the results show that 1. the error in estimating the respiratory rate increases as a function of the time spent in AF, 2. the leads farthest away from the atria, i.e., V4, V5, V6, exhibit the best performance due to lower f-wave amplitudes, 3. lower errors in leads with sim- ilar f-wave amplitude are due to a more pronounced respi- ratory influence, and 4. the respiratory influence is higher in V2, V3, and V4 compared to other precordial leads.