The aim of this study was to evaluate six peak flow meters, comparing them to a flow signal recorded under laboratory and clinical conditions (LC and CC, respectively). The six peak flow meters studied were the PF-Control (A), the Personal Best (Healthscan) (B), the logarithmic scale and linear Vitalographs (Vitalograph) (C and D), the Miniwright (Airmed) (E), and the Assess (Healthscan) (F), LC: Readings were compared to those obtained with a computer-controlled syringe attached to a servomechanism (Pulmonary Waveform Generator System) at flow levels of 185, 302 and 595 L/m before and after clinical measurements. CC: Readings from each model, taken randomly and consecutively, were compared with those obtained by pneumotachometer in 50 individuals with different air flow levels (range 86 to 888 L/min), ending by repeating the measurement procedure with the first device used. The LC results were as follows: precision varied among the six models from 0.9% (F) to 9.4% (E), and in no case was there a statistically significant difference from the servosyringe reading; reproducibility (coefficient of variation) ranged from 0.7% (F) to 3.3% (B); stability after clinical testing showed results that differed from those of the servosyringe in a range of -18.3 L/min (A) to +16.7 L/min (B) and the difference was significant only for device B. The CC results were as follows: in comparison with the pneumotachometer readings, percent error was within NHLBI limits for devices A and B in excess of the limits for the other peak flow meters. Analysis of concordance showed that differences were related to level of flow for all devices except F and that measurement errors conformed to an equation of linear regression. We conclude that all the studied models give readings that are biased in direct proportion yo the flow measured. Precision was good and the devices remained stable over more than 200 readings, such that peak flow meters are appropriate for repeated measurements in a single patient.