OBJECTIVE--To define the ability of analysis of the signal averaged P wave to identify patients with paroxysmal atrial fibrillation (AF) and establish whether differences in quantitative variables between patients and controls are due to concurrent cardiopulmonary disease, greater atrial dimension, or to unrelated changes in atrial electrophysiology. DESIGN--An observational parallel group study. SETTING--Cardiac department of a busy district general hospital. PATIENTS--58 participants without cardiopulmonary disease (24 with paroxysmal AF and 34 controls, group A) and 57 with cardiac or respiratory conditions (31 with paroxysmal AF and 26 controls, group B). Mean (range) age of patients was 54 (25-71) and controls 53 (34-78) for group A and 65 (45-81) and 62 (36-78) respectively for group B. Left atrial size was similar in patients and controls in each group (mean (SEM)) group A: 2.39 (0.1) v 2.19 (0.07) cm; group B: 2.51 (0.10) v 2.71 (0.12) cm). MAIN OUTCOME MEASURES--Analysis of the P wave after P-wave-specific signal averaging. Filtered P wave duration and spatial velocity were calculated. Energies contained in frequency bands from 20, 30, 40, 60, and 80 to 150 Hz after spectral analysis were expressed as absolute values (P20, P30 etc) and ratios of high to low frequency energy (PR20, PR30, etc). RESULTS--Duration and peak spatial velocity were increased in patients with paroxysmal AF (median (interquartile range) duration group A: 144 (137-155) v 136 (129-143) ms, P = 0.007; group B: 155 (144-159) v 142 (136-151) ms, P = 0.002; peak spatial velocity group A: 16.5 (14.1-21.2) v 14.5 (11.7-18.1) mV/s, P = 0.02; group B: 18.9 (14.8-21.8) v 14.3 (12.6-17.6) mV/s, P = 0.01). Energy contained in frequency bands from 20, 30, 40, 60 and 80 to 150 Hz was expressed as absolute values (P20, P30, P40, P60, and P80) and percentage energy ratios. P30, P60, and P80 were significantly greater in patients with AF in group A (for example P60: 3.9 (3.0-5.3) v 3.1 (2.0-4.3) microV2.s, P = 0.02) and P20, P30, and P40 were increased in those with AF in group B (for example P40: 16.7 (9.9-20.8) v 10.8 (8.1-14.8) microV2.s, P = 0.02). A score developed from logistic regression analysis of duration and P60 identified patients with paroxysmal AF with a sensitivity of 81% and specificity of 73%. CONCLUSIONS--Increased P wave duration and magnitude are associated with paroxysmal AF with and without additional cardiopulmonary disease. The discriminant ability of the signal averaged P wave is improved by analysis of duration and a magnitude variable. These results invite prospective assessment of the ability of the signal averaged P wave to predict paroxysmal AF in unselected patients.