We studied 40 patients with chronic obstructive pulmonary disease (COPD) to determine whether measurements of pulmonary function could predict a fall in arterial oxygen pressure (PaO2) with exercise. The PaO2 fell more than 3 mm Hg in 21 patients (group 1), did not change (+/- 3 mm Hg) in nine patients (group 2), and increased more than 3 mm Hg in ten patients (group 3). Group 3 had significantly less severe expiratory obstruction than groups 1 and 2. The most significant variables in predicting a change in PaO2 with exercise were the ratio of the forced expiratory volume in one second over the forced vital capacity (FEV1/FVC) and the single-breath carbon monoxide diffusing capacity (Dsb). Measurements of FEV1/FVC of 0.50 or more and Dsb of 20 ml/min/mm Hg or more were 100 percent predictive in excluding a fall in PaO2 with exercise. Measurements below these thresholds could not be used reliably to predict which patients would develop worsening hypoxemia with exercise. Because of wide variability in reference values from eight different published studies for diffusing capacity, recommended criteria based on the percent predicted Dsb should be used with caution. We conclude that pulmonary function measurements cannot be used to predict exercise-induced hypoxemia in patients with COPD; however, the measurements may be useful in identifying patients whose condition is less severe who are unlikely to develop worsening hypoxemia with exercise.