We assessed (1) the sensitivity and specificity of exercise oxygen saturation measurement (EOS) for the diagnosis of Pneumocystis carinii pneumonia (PCP); and (2) the cost of introducing this indirect diagnostic test compared with that of standard diagnostic strategies for PCP. In a prospective study, 85 HIV-infected patients with suspected PCP underwent EOS, followed by induced sputum (IS) and bronchoalveolar lavage (BAL) if IS was negative for P. carinii. The prevalence of PCP was 0.22, the sensitivity of IS was 0.6, and its specificity was perfect. The cost ratios of IS to BAL and EOS to BAL were 0.1 and 0.2, respectively. A desaturation of three points was the best cutoff point, giving perfect sensitivity and a specificity of 0.77. The cost analysis showed that the introduction of EOS into diagnostic strategies for PCP is highly justified when the local prevalence is low. Exercise oxygen saturation measurement is simple and safe, and the results are available rapidly; its sensitivity is perfect and its specificity good. Its economic utility depends on its cost and the local prevalence of PCP in the test population.