We have compared the pacing rate responses during cardiopulmonary exercise testing in 11 patients (mean 59 years, six female) with implanted QT sensing rate adaptive pacemakers who were randomly programmed to 1-month periods in the linear and nonlinear algorithms using a double-blind, cross-over design. Exercise testing was performed at the end of each month block and symptoms were scored with the MacMaster questionnaire. With exercise, the time to a 10 beats/min increment in rate was significantly less with the nonlinear compared to the linear algorithm (126 sec vs 255 sec, P = 0.02) but there were no significant differences in exercise duration, the peak pacing rate, the peak VO2, the VO2 at the anaerobic threshold or the mean correlation coefficients of the pacing rate VO2 relationship. Rate oscillation occurred in seven patients in the linear algorithm and in two patients in the nonlinear setting. Initial deceleration of the pacing rate at the onset of exercise occurred in seven patients in the linear algorithm and in four patients in the nonlinear setting. The nonlinear algorithm is associated with a faster response time during exercise and fewer instances of rate instability. However, it has not overcome the problem of a dip in the pacing rate at the beginning of exercise. The major difference in the function of the two algorithms is faster initial acceleration with the nonlinear algorithm. This is explained by the significantly higher values of the slope setting at the lower rate limit for the nonlinear versus the linear algorithm (6.3 ms/ms vs 5.1 ms/ms).