Automatic mode switching (AMS) prevents tracking of paroxysmal atrial fibrillation (AF) in dual chamber pacing. The correct detection of AF can be affected by the programmed atrial sensitivity (AS). We prospectively studied the relationship between AS, AF undersensing, and AMS, using unfiltered bipolar intracardiac atrial electrograms recorded from 17 patients during sinus rhythm (SR) and in AF. Overall, 780 rhythms were recorded and replayed onto three dual chamber pacemaker models using different AMS algorithms (Thera DR 7940, Marathon DDDR 294-09, and Meta DDDR 1254), and the ventricular responses were measured. AS was randomly programmed in steps from the highest available AS to half of the mean atrial P wave amplitude (PWA), and the percentage of appropriate AMS responses (defined as a ventricular pacing rate at the expected AMS mode) were recorded. AMS efficacy was related to the programmed AS settings in an exponential manner. At low AS settings, a higher percentage of tests were associated with absence of, or with intermittent AMS and tracking of AF, whereas at higher AS, oversensing of noise during SR occurred. An optimal AS measured approximately 1.3 mV, representing about one-third of the PWA measured during SR, although oversensing of SR and undersensing of AF continued to occur in 14% of tests and time, respectively, due to the high variation in PWA during AF. Thus, a fixed AS cannot eliminate AF undersensing without inviting noise oversensing, suggesting the need for automatic adjustments of AS, or the use of a rate-limiting algorithm to prevent rate oscillation during intermittent AF sensing. In conclusion, AMS functions of existing pacemakers were significantly limited by the undersensing of AF and oversensing of noise. Proper adjustment of the AS is important to enable effective AMS during AF.