To evaluate the reliability of noninvasive oximeters during exercise in patients with cystic fibrosis (CF), we studied 9 patients during exercise tests, using 3 devices (Hewlett-Packard 47201A, Biox-Ohmeda 3700, Nellcor 100), comparing oximetry readings with arterial blood co-oximetry. Fifty samples from preexercise, exercise, and recovery were collected; and paired measurements were analyzed for bias (arterial blood gas SaO2--noninvasive SaO2) and precision (1 SD of the bias). Values were available for the Hewlett-Packard and the Biox-Ohmeda at all 50 collection times. The Nellcor failed to read on four occasions, all during exercise. The accuracy of the Biox-Ohmeda readings was significantly worse during exercise than rest, while the Nellcor and the Hewlett-Packard performance did not differ significantly between exercise and rest. During hypoxemia (SaO2 < 90 percent as determined by co-oximetry on arterial blood), the Biox-Ohmeda and Nellcor consistently overestimated saturation, with mean bias of -6.1 percent and -3.0 percent, respectively. The Hewlett-Packard was more reliable and valid during hypoxemia, with a mean bias of -1.0 percent. The Nellcor read significantly more accurately in the six patients with severe digital clubbing than in the three patients with mild digital clubbing. It is important to be aware of possible limitations of noninvasive oximeters before accepting them as accurate.