Accurate detection of atrial fibrillation (AF) is essential for appropriate operation of an implantable atrial defibrillator (IAD). However, during episodes of sinus tachycardia, distinction between AF and sinus rhythm (SR) using the "quiet interval" and "baseline crossing" analysis in the detection algorithm of the IAD may be difficult. The efficacy of this AF detection algorithm was tested in five patients implanted with an IAD (METRIX, Model 3000 or 3020, InControl Inc.) during treadmill exercise testing. The IADs were programmed to Monitor Mode with a wake up cycle of 1 minute for AF detection using the device nominal parameters or modified parameters, and to mark rhythms appropriate for shock delivery. A mean peak heart rate of 137 +/- 26 beats/min was reached during maximum exercise, and one patient developed transient AF. Seventy-eight (75 in SR, 3 in AF) and 91 (89 in SR, 2 in AF) runs of AF detection were performed using the nominal and modified parameters, respectively. The IAD detected AF and SR accurately, except for one episode of false-positive AF detection during sinus tachycardia at the nominal settings, but inappropriate shocks were prevented by minimum RR interval criteria that limited discharge at high heart rate. These results indicate that the AF detection algorithm in the IAD may become more vulnerable to false-positive AF detection during sinus tachycardia, which were avoided by reprogramming the Quiet Interval and minimum RR interval criteria for AF detection. Exercise testing appeared useful to program optimal settings of the IAD in preparation for daily activities.